ContextStream MCP Server

officiel

Mémoire persistante et recherche sémantique pour les assistants de codage IA à travers les sessions.

Documentation

Docs · Parcourir les sections

01 · Assistant de configuration

Une commande, entièrement configuré.

Exécutez une commande pour vous authentifier (connexion navigateur/appareil), créer une clé API, générer des règles et écrire la configuration MCP correcte pour votre outil (y compris le schéma servers de VS Code).

terminal · terminal · macOS / Linux

curl -fsSL https://contextstream.io/scripts/mcp.sh | bash

terminal · powershell · Windows

irm https://contextstream.io/scripts/mcp.ps1 | iex

Déjà installé ? Relancez l'assistant de configuration

terminal · terminal · relancer

contextstream-mcp setup

Ce qu'il fait : écrit la configuration MCP (VS Code servers, Cursor/Cline/etc mcpServers), génère des règles (Standard/Améliorées) et peut lier des projets à un espace de travail.

Boîte à outils consolidée : L'assistant configure par défaut environ 11 outils de domaine consolidés (réduction d'environ 75 % des tokens). Optionnel : mode router (environ 2 méta-outils, le plus compact). Voir le catalogue complet des outils.

Prévisualisez les modifications sans écrire de fichiers : contextstream-mcp setup --dry-run

Flux de travail en équipe

Mémoire partagée, compétences et mise en contexte.

Les comptes d'équipe offrent plus que des projets partagés. Pendant contextstream-mcp setup, l'assistant détecte la capacité d'équipe et présente des conseils sur l'espace de travail. Dans votre éditeur, chaque appel session(action="context") peut inclure des recommandations d'équipe, des indices de gouvernance, des signaux de priorité et des artefacts liés.

Choisir l'espace de travail partagé

Liez chaque dépôt à l'espace de travail d'équipe lors de la configuration afin que l'indexation, les décisions et les tickets restent alignés.

Partager les compétences d'équipe

skill(action="share", scope="team") publie des flux de travail réutilisables que les coéquipiers associent automatiquement dans session(action="context").

Changer la portée d'exécution

Les comptes à double contexte utilisent --account-mode=team|personal|auto ou session set_account_mode dans MCP.

Suivre le travail avec des entités

Les tickets, transferts, incidents et versions utilisent des références indexées — durables à travers les sessions et les coéquipiers.

La télécommande hébergée est le mode par défaut. Le binaire MCP local est uniquement pour la récupération — définissez CONTEXTSTREAM_ALLOW_LOCAL_MCP=1 lorsque cela est explicitement nécessaire. Voir configuration d'équipe pour les invitations/rôles et la liste de contrôle post-connexion.

Raccourcis CLI

Commandes non interactives (CI et rafraîchissement).

Exécutez-les sans l'assistant interactif — idéal après les mises à niveau, l'intégration d'équipe, la rotation des informations d'identification ou les changements d'espace de travail. Également présenté dans contextstream-mcp --help.

Commande	Quand l'utiliser
contextstream-mcp update-hooks --scope=global	Après une mise à niveau ou l'adhésion à un espace de travail d'équipe — rafraîchir les hooks PreToolUse/UserPromptSubmit.
contextstream-mcp update-rules --scope=all	Régénérer .cursorrules / CLAUDE.md / AGENTS.md avec les dernières directives de flux de travail d'équipe.
contextstream-mcp update-configs --scope=global	Réécrire les configurations MCP après des changements de clé API ou d'espace de travail.
contextstream-mcp migrate-remote --scope=all	Convertir les anciennes configurations stdio locales en transport distant hébergé.
contextstream-mcp detect-editors --format=json	Script pour détecter les éditeurs installés (amorçage/CI).
contextstream-mcp generate-configs --transport=remote --preauth	Émettre des charges utiles de configuration JSON sans écrire de fichiers.
contextstream-mcp configure --transcripts=on --scope=all	Définir les valeurs par défaut de capture de transcription de manière non interactive.

terminal · mode compte · équipe vs personnel

# Default: follow account (auto)
contextstream-mcp --account-mode=auto

# Force team-scoped reads/writes
contextstream-mcp --account-mode=team

# Or set once in shell profile:
export CONTEXTSTREAM_ACCOUNT_MODE=team

02 · Configuration manuelle

Configurations par client.

Utilisez le format correct pour chaque client (VS Code utilise servers ; de nombreux autres clients utilisent mcpServers).

Boîte à outils consolidée : Par défaut, le serveur expose environ 11 outils de domaine consolidés (réduction d'environ 75 % des tokens par rapport aux anciens outils granulaires). Pour encore moins d'outils, ajoutez "CONTEXTSTREAM_PROGRESSIVE_MODE": "true" au bloc env pour environ 2 méta-outils de routage. Voir le catalogue complet des outils.

Aller à votre outil

Cursor / VS Code Windsurf Codex CLI OpenCode CLI Claude Code Claude Desktop Cline Kilo Code Roo Code Antigravity

Qu'est-ce que MCP ?

Un protocole ouvert pour l'IA.

Le Model Context Protocol (MCP) est un standard ouvert qui permet aux assistants IA de se connecter à des outils et sources de données externes. Avec le serveur MCP de ContextStream, vos outils IA peuvent :

Se souvenir des conversations et des décisions à travers les sessions
Rechercher dans votre base de code et votre documentation de manière sémantique
Construire et interroger des graphes de connaissances
Partager le contexte entre différents outils IA

Langage naturel

Demandez simplement. L'IA gère les outils.

Vous n'avez pas besoin de mémoriser les noms des outils ni de les appeler directement. Décrivez simplement ce que vous voulez en langage clair et votre assistant IA utilisera automatiquement les bons outils.

Demandez naturellement

· "résumé de session"
· "qu'avons-nous décidé à propos de l'authentification ?"
· "rappelle-toi que nous utilisons PostgreSQL"
· "recherche le code de paiement"

L'IA gère le reste

· Trouve automatiquement le contexte pertinent
· Rappelle les décisions passées si nécessaire
· Sauvegarde les informations importantes en mémoire
· Recherche dans le code et la documentation

Exemple · "résumé de session"

Natural language example: typing 'session summary' and the AI automatically uses context_smart

L'IA comprend votre intention et appelle les outils ContextStream appropriés en arrière-plan.

Prérequis

Ce dont vous avez besoin.

Un compte ContextStream (l'assistant de configuration peut créer une clé API via la connexion navigateur).

Enrichir le contexte

Intégrations GitHub + Slack

MCP donne à votre IA une mémoire persistante. Connecter GitHub et Slack rend cette mémoire beaucoup plus riche — votre IA peut automatiquement référencer les PR, les tickets et les discussions d'équipe lorsqu'elle répond aux questions.

Enrichissement automatique du contexte

Lorsque vous appelez context_smart ou session_smart_search, les tickets GitHub, PR et discussions Slack pertinents sont automatiquement inclus. Aucun outil supplémentaire nécessaire.

GitHubSync tickets, PR, versions et commentaires. Les décisions sont automatiquement extraites des discussions.SlackSync canaux et fils de discussion. Les conversations à fort engagement sont notées et priorisées.

Exemples de prompts

· "Qu'avons-nous décidé à propos de l'authentification ?" — trouve les décisions des tickets GitHub + fils Slack
· "Montre-moi l'activité récente sur le système de paiement" — présente les PR, tickets et discussions d'équipe
· "Quelles leçons avons-nous tirées de la dernière panne ?" — récupère les informations de Slack et GitHub
· "Donne-moi un résumé hebdomadaire de l'activité GitHub" — utilise integration(provider="github", action="summary", ...)
· "Recherche dans toutes les intégrations les discussions sur la migration de base de données" — utilise integration(provider="all", action="search", ...)
· "Donne-moi un résumé hebdomadaire d'équipe toutes sources confondues" — utilise integration(provider="all", action="summary", ...)

Référence rapide des actions des outils d'intégration

Utilisez integration(provider="github|slack|all", action="...")

GitHub (provider="github")

action="stats" — Statistiques et état de synchronisation
action="search" — Recherche avec filtres d'état/période
action="activity" — Flux d'activité (filtre jours)
action="knowledge" — Décisions/leçons extraites
action="summary" — Résumé hebdomadaire/mensuel
action="repos" — Lister les dépôts synchronisés
action="issues" — Lister les tickets/PR

Slack (provider="slack")

action="stats" — Statistiques et état de synchronisation
action="search" — Recherche avec filtres de canal/période
action="discussions" — Fils à fort engagement
action="knowledge" — Décisions/leçons extraites
action="summary" — Résumé hebdomadaire/mensuel
action="channels" — Lister les canaux synchronisés

Multi-source (provider="all")

action="status" — Vérifier l'état de synchronisation et la santé de toutes les intégrations connectées
action="search" — Rechercher dans toutes les intégrations connectées en une seule requête
action="summary" — Résumé d'activité unifié toutes sources (filtre jours)
action="knowledge" — Obtenir les décisions, leçons et informations de toutes les sources

Client · Cursor / VS Code

Cursor / VS Code

Cursor et VS Code utilisent des schémas de configuration MCP différents. Cursor utilise toujours un processus MCP local, mais VS Code/Copilot peut désormais utiliser le MCP ContextStream hébergé directement via HTTP.

terminal · .cursor/mcp.json (projet) ou ~/.cursor/mcp.json (global)

{
  "mcpServers": {
    "contextstream": {
      "command": "contextstream-mcp",
      "env": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "your_api_key"
      }
    }
  }
}

Recommandé pour VS Code / Copilot : installation distante en un clic

Installez le MCP ContextStream hébergé et laissez VS Code gérer OAuth lors de la première utilisation. Aucun binaire local ni clé API ne doit résider dans .vscode/mcp.json.

Installer dans VS Code Documentation MCP VS Code

terminal · .vscode/mcp.json (MCP natif VS Code, distant)

{
  "servers": {
    "contextstream": {
      "type": "http",
      "url": "https://mcp.contextstream.io/mcp?default_context_mode=fast"
    }
  }
}

Vous préférez la ligne de commande ? Ajoutez le serveur distant avec code --add-mcp

terminal · terminal

code --add-mcp '{"name":"contextstream","type":"http","url":"https://mcp.contextstream.io/mcp?default_context_mode=fast"}'

Lors de la première utilisation, VS Code devrait demander l'autorisation de ContextStream, puis terminer la configuration automatiquement.

Auto-hébergé ? Pointez la même configuration distante vers votre propre URL de passerelle MCP au lieu de https://mcp.contextstream.io/mcp?default_context_mode=fast.

Client · OpenCode CLI

OpenCode CLI

Pour utiliser ContextStream avec OpenCode CLI, ajoutez la configuration du serveur MCP à votre fichier ~/.config/opencode/opencode.json (ou opencode.json à la racine de votre projet) :

terminal · ~/.config/opencode/opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "contextstream": {
      "type": "local",
      "command": ["contextstream-mcp"],
      "enabled": true,
      "environment": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "your_api_key"
      }
    }
  }
}

Après avoir modifié la configuration, redémarrez OpenCode pour qu'il puisse charger le serveur MCP ContextStream.

Client · Codex CLI

Codex CLI

Pour utiliser ContextStream avec Codex CLI, ajoutez la configuration du serveur MCP à votre fichier ~/.codex/config.toml :

terminal · ~/.codex/config.toml

[mcp_servers.contextstream]
command = "contextstream-mcp"
args = []

[mcp_servers.contextstream.env]
CONTEXTSTREAM_API_URL = "https://api.contextstream.io"
CONTEXTSTREAM_API_KEY = "your_api_key"

Après avoir modifié la configuration, redémarrez Codex pour qu'il puisse charger le serveur MCP ContextStream.

Client · Claude Code

Claude Code (CLI)

Ajoutez ContextStream à Claude Code en exécutant cette commande depuis le répertoire de votre projet :

terminal · terminal

claude mcp add --transport stdio contextstream --env CONTEXTSTREAM_API_URL=https://api.contextstream.io --env CONTEXTSTREAM_API_KEY=your_api_key -- contextstream-mcp

Mise en garde Windows (Windows natif, pas WSL) : utilisez cmd /c contextstream-mcp après --.

Alternative : add-json (stdio)

terminal · terminal · add-json

claude mcp add-json contextstream \
'{"type":"stdio","command":"contextstream-mcp","args":[],"env":{"CONTEXTSTREAM_API_URL":"https://api.contextstream.io","CONTEXTSTREAM_API_KEY":"your_api_key"}}'

Conseil : pour les configurations d'équipe, préférez un fichier .mcp.json validé (portée projet) plutôt que d'intégrer des clés dans l'historique du shell.

Cela ajoute ContextStream à ~/.claude.json sous le chemin de votre projet, le rendant disponible lorsque vous travaillez dans ce répertoire.

Options de portée MCP

· Local (par défaut) : Privé pour vous, projet actuel uniquement → stocké dans ~/.claude.json sous le chemin du projet.
· Utilisateur (--scope user) : Privé pour vous, tous les projets → stocké dans ~/.claude.json globalement.
· Projet (--scope project) : Partagé avec l'équipe → stocké dans .mcp.json à la racine du projet (valider dans git).

Pour le partage d'équipe, utilisez la portée projet pour créer un fichier .mcp.json qui peut être validé dans git :

terminal · .mcp.json (portée projet)

{
  "mcpServers": {
    "contextstream": {
      "command": "contextstream-mcp",
      "env": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "your_api_key"
      }
    }
  }
}

Après avoir ajouté le serveur MCP, redémarrez Claude Code pour que les modifications prennent effet. Vérifiez que le serveur est chargé avec claude mcp list. Pour les serveurs à portée projet provenant de .mcp.json, Claude Code demandera une approbation lors de la première utilisation.

Client · Claude Desktop

Claude Desktop (application GUI)

Ajoutez ContextStream à l'application Claude Desktop :

macOS : ~/Library/Application Support/Claude/claude_desktop_config.json

Windows : %APPDATA%\Claude\claude_desktop_config.json

terminal · claude_desktop_config.json

{
  "mcpServers": {
    "contextstream": {
      "command": "contextstream-mcp",
      "env": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "your_api_key"
      }
    }
  }
}

Après avoir modifié la configuration, quittez et redémarrez Claude Desktop pour que les modifications prennent effet.

Client · Windsurf

Windsurf

Ajoutez ContextStream à Windsurf en modifiant le fichier de configuration MCP global :

Config : ~/.codeium/windsurf/mcp_config.json

terminal · ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "contextstream": {
      "command": "contextstream-mcp",
      "env": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "your_api_key"
      }
    }
  }
}

Fichiers de règles

· Global : ~/.codeium/windsurf/memories/global_rules.md
· Projet : .windsurf/rules/contextstream.md

Windsurf prend en charge les hooks pour l'application automatique des règles ContextStream. Après avoir modifié la configuration, redémarrez Windsurf pour que les modifications prennent effet.

Client · Cline

Cline

Ajoutez ContextStream à votre configuration MCP Cline. Cliquez sur l'icône MCP Servers dans Cline, sélectionnez l'onglet "Configure", puis cliquez sur "Configure MCP Servers" pour modifier :

terminal · cline_mcp_settings.json

{
  "mcpServers": {
    "contextstream": {
      "command": "contextstream-mcp",
      "env": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "your_api_key"
      }
    }
  }
}

Après avoir modifié la configuration, redémarrez Cline pour que les modifications prennent effet. Vous pouvez également utiliser alwaysAllow pour approuver automatiquement des outils spécifiques.

Client · Kilo Code

Kilo Code

Ajoutez ContextStream à votre configuration MCP Kilo Code. Vous pouvez configurer les serveurs MCP globalement ou par projet :

Global : Cliquez sur Paramètres → MCP Servers → Installed → Edit Global MCP pour ouvrir mcp_settings.json

Projet : .kilocode/mcp.json à la racine de votre projet

terminal · .kilocode/mcp.json (ou mcp_settings.json)

{
  "mcpServers": {
    "contextstream": {
      "command": "contextstream-mcp",
      "env": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "your_api_key"
      }
    }
  }
}

Les configurations au niveau projet prévalent sur les configurations globales. Redémarrez Kilo Code après modification.

Client · Roo Code

Roo Code

Ajoutez ContextStream à votre configuration MCP Roo Code. Vous pouvez configurer les serveurs MCP globalement ou par projet :

Global : Cliquez sur l'icône des paramètres → Edit Global MCP pour ouvrir mcp_settings.json

Projet : .roo/mcp.json à la racine de votre projet

terminal · .roo/mcp.json (ou mcp_settings.json)

{
  "mcpServers": {
    "contextstream": {
      "command": "contextstream-mcp",
      "env": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "your_api_key"
      }
    }
  }
}

Les configurations au niveau projet prévalent sur les configurations globales. Redémarrez Roo Code après modification.

Client · Antigravity

Antigravity (Google)

Antigravity utilise des fichiers .mcp.json à portée projet avec le même format que Cursor/Claude Desktop :

terminal · .mcp.json (racine du projet)

{
  "mcpServers": {
    "contextstream": {
      "command": "contextstream-mcp",
      "env": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "your_api_key"
      }
    }
  }
}

Utilisateurs Windows : utilisez un wrapper cmd

terminal · .mcp.json (Windows)

{
  "mcpServers": {
    "contextstream": {
      "command": "cmd",
      "args": ["/c", "contextstream-mcp"],
      "env": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "your_api_key"
      }
    }
  }
}

Fichiers de règles

· Global : ~/.gemini/GEMINI.md
· Espace de travail : .agent/rules/contextstream.md

Accédez via le menu "..." → "MCP Servers" → "View raw config" pour vérifier votre configuration. Redémarrez Antigravity après modification.

Règles d'éditeur

Améliorer l'utilisation automatique de ContextStream

Recommandé

La fonctionnalité de contexte automatique de ContextStream charge le contexte de l'espace de travail automatiquement lorsque vous utilisez un outil ContextStream. Cependant, les assistants IA peuvent ne pas toujours sauvegarder proactivement les décisions ou rappeler le contexte passé. L'ajout de règles d'IA d'éditeur améliore la cohérence et garantit que l'IA capture automatiquement les décisions, les préférences et le contexte important tout au long de vos conversations.

Critique : conventions de nommage des outils MCP

Différents outils IA utilisent différentes conventions de nommage pour les outils MCP. Utiliser le mauvais format empêchera les outils d'être trouvés.

Outil IA	Format	Exemple
Claude Code	mcp____	mcp__contextstream__session_init
Codex CLI / OpenCode CLI	(nom brut)	session_init
Cursor / Windsurf / Cline	(nom brut)	session_init
Kilo Code / Roo Code	(nom brut)	session_init

Résumé : Seul Claude Code utilise le préfixe mcp__contextstream__. Tous les autres outils utilisent les noms d'outils bruts.

Vous pouvez ajouter des règles ContextStream à deux niveaux : Global (s'applique à tous les projets) ou Projet (s'applique à un seul projet).

Règles globales (tous les projets)

Ajoutez-les une fois et elles s'appliqueront automatiquement à chaque projet :

Éditeur	Emplacement des règles globales
Cursor	Paramètres → Général → Règles pour l'IA
Windsurf	~/.codeium/windsurf/memories/global_rules.md
Cline	~/Documents/Cline/Rules/
Kilo Code	~/.kilocode/rules/
Roo Code	~/.roo/rules/
Claude Code	~/.claude/CLAUDE.md
Codex CLI	~/.codex/AGENTS.md (global) ou dossier parent (ex. ~/dev/AGENTS.md)
OpenCode CLI	~/.config/opencode/AGENTS.md

Règles de projet (projet unique)

Ajoutez-les à un projet spécifique. Pour les règles basées sur des dossiers Cursor, définissez le mode d'activation sur "Toujours actif" afin que les règles soient toujours actives.

Éditeur	Emplacement des règles de projet
Cursor	.cursorrules ou .cursor/rules/*.mdc
Windsurf	.windsurf/rules/contextstream.md
Cline	fichier .clinerules ou dossier .clinerules/
Kilo Code	.kilocode/rules/
Roo Code	.roo/rules/contextstream.md ou dossier .roo/rules/
Claude Code	CLAUDE.md à la racine du projet
Codex CLI	AGENTS.md à la racine du projet
OpenCode CLI	AGENTS.md à la racine du projet
Aider	.aider.conf.yml à la racine du projet

Modes d'activation (Cursor, Kilo Code & Roo Code)

Lors de l'utilisation de règles basées sur des dossiers (ex., .cursor/rules/), chaque fichier de règles a un mode d'activation :

Toujours actif — Toujours actif (recommandé pour ContextStream)
Manuel — Uniquement lorsque vous @mentionnez la règle
Décision du modèle — L'IA décide en fonction de la description
Glob — Actif pour les motifs de fichiers correspondants

Les règles globales et les fichiers au niveau racine (.cursorrules) sont toujours actifs.

Vous préférez l'assistant de configuration ? Exécutez-le d'abord. Sinon, ajoutez ces règles standard manuellement. Exemples pour chaque éditeur :

Claude Code

Créez un fichier CLAUDE.md à la racine de votre projet ou ajoutez à votre ~/.claude/CLAUDE.md global :

terminal · CLAUDE.md (Standard)

## 🚨 MANDATORY RULE: DISPLAY AT START OF EVERY RESPONSE 🚨

<contextstream_rules>
1. **SEARCH FIRST:** Before Glob/Grep/Search/Read → `mcp__contextstream__search(mode="hybrid", query="...")` FIRST
2. **EVERY MESSAGE:** Call `mcp__contextstream__context_smart(user_message="...")` at start of response
3. **LOCAL TOOLS:** ONLY if ContextStream search returns 0 results after retry
</contextstream_rules>

**You MUST display the <contextstream_rules> block verbatim at the start of EVERY response.**
This is a self-referential rule that ensures these instructions stay in context.

---

## ContextStream v0.4.x (Hooks Enforced)

Rules Version: 0.4.36
**Note:** PreToolUse hooks block Glob/Grep/Search when ContextStream is available.

### Required Every Message

| Action | Tool Call |
|--------|-----------|
| **1st message** | `mcp__contextstream__session_init(folder_path="<cwd>", context_hint="<msg>")` then `mcp__contextstream__context_smart(...)` |
| **2nd+ messages** | `mcp__contextstream__context_smart(user_message="<msg>", format="minified", max_tokens=400)` |
| **Code search** | `mcp__contextstream__search(mode="hybrid", query="...")` — BEFORE any local tools |
| **Save decisions** | `mcp__contextstream__session(action="capture", event_type="decision", ...)` |

### Search Modes

| Mode | Use Case |
|------|----------|
| `hybrid` | General code mcp__contextstream__search (default) |
| `keyword` | Exact symbol/string match |
| `exhaustive` | Find ALL matches (grep-like) |
| `semantic` | Conceptual questions |

### Why ContextStream First?

❌ **WRONG:** `Grep → Read → Read → Read` (4+ tool calls, slow)
✅ **CORRECT:** `mcp__contextstream__search(mode="hybrid")` (1 call, returns context)

ContextStream search is **indexed** and returns semantic matches + context in ONE call.

### Quick Reference

| Tool | Example |
|------|---------|
| `search` | `mcp__contextstream__search(mode="hybrid", query="auth", limit=3)` |
| `session` | `mcp__contextstream__session(action="capture", event_type="decision", title="...", content="...")` |
| `memory` | `mcp__contextstream__memory(action="list_events", limit=10)` |
| `graph` | `mcp__contextstream__graph(action="dependencies", file_path="...")` |

### Lessons (Past Mistakes)

- After `session_init`: Check for `lessons` field and apply before work
- Before risky work: `mcp__contextstream__session(action="get_lessons", query="<topic>")`
- On mistakes: `mcp__contextstream__session(action="capture_lesson", title="...", trigger="...", impact="...", prevention="...")`

### Plans & Tasks

When user asks for a plan, use ContextStream (not EnterPlanMode):
1. `mcp__contextstream__session(action="capture_plan", title="...", steps=[...])`
2. `mcp__contextstream__memory(action="create_task", title="...", plan_id="<id>")`

Full docs: https://contextstream.io/docs/mcp/tools

Afficher les règles améliorées (verbeuses)

terminal · CLAUDE.md (Améliorées)

## 🚨 MANDATORY RULE: DISPLAY AT START OF EVERY RESPONSE 🚨

<contextstream_rules>
1. **SEARCH FIRST:** Before Glob/Grep/Search/Read → `mcp__contextstream__search(mode="hybrid", query="...")` FIRST
2. **EVERY MESSAGE:** Call `mcp__contextstream__context_smart(user_message="...")` at start of response
3. **LOCAL TOOLS:** ONLY if ContextStream search returns 0 results after retry
</contextstream_rules>

**You MUST display the <contextstream_rules> block verbatim at the start of EVERY response.**
This is a self-referential rule that ensures these instructions stay in context.

---

## ContextStream v0.4.x Integration (Enhanced)

You have access to ContextStream MCP tools for persistent memory and context.
v0.4.x uses **~11 consolidated domain tools** for ~75% token reduction vs previous versions.
Rules Version: 0.4.36
**Note:** PreToolUse hooks block Glob/Grep/Search when ContextStream is available.

## TL;DR - REQUIRED EVERY MESSAGE

| Message | What to Call |
|---------|--------------|
| **1st message** | `mcp__contextstream__session_init(folder_path="...", context_hint="<user's message>")`, then `mcp__contextstream__context_smart(user_message="<user's message>", format="minified", max_tokens=400)` |
| **2nd+ messages** | `mcp__contextstream__context_smart(user_message="<user's message>", format="minified", max_tokens=400)` |
| **🔍 ANY code search** | `mcp__contextstream__search(mode="hybrid", query="...")` — ALWAYS before Glob/Grep/Search/Read |
| **Before risky/non-trivial work** | `mcp__contextstream__session(action="get_lessons", query="<topic>")` |
| **After completing task** | `mcp__contextstream__session(action="capture", event_type="decision", ...)` - MUST capture |
| **User frustration/correction** | `mcp__contextstream__session(action="capture_lesson", ...)` - MUST capture lessons |

**NO EXCEPTIONS.** Do not skip even if you think you have enough context.

**First message rule:** After `session_init`, always call `context_smart` before any other tool or response.

**Context Pack (Pro+):** If enabled, use `mcp__contextstream__context_smart(..., mode="pack", distill=true)` for code/file queries. If unavailable or disabled, omit `mode` and proceed with standard `context_smart` (the API will fall back).

**Tool naming:** Use the exact tool names exposed by your MCP client. Claude Code typically uses `mcp__<server>__<tool>` where `<server>` matches your MCP config (often `contextstream`). If a tool call fails with "No such tool available", refresh rules and match the tool list.

---

## Consolidated Domain Tools Architecture

v0.4.x consolidates ~58 individual tools into ~11 domain tools with action/mode dispatch:

### Standalone Tools (Always Call)
- **`session_init`** - Initialize session with workspace detection + context
- **`context_smart`** - Semantic search for relevant context (CALL EVERY MESSAGE, including immediately after `session_init`)

### Domain Tools (Use action/mode parameter)

| Domain | Actions/Modes | Example |
|--------|---------------|---------|
| **`search`** | mode: semantic, hybrid, keyword, pattern, exhaustive, refactor | `mcp__contextstream__search(mode="hybrid", query="auth implementation", limit=3)` |
| **`session`** | action: capture, capture_lesson, get_lessons, recall, remember, user_context, summary, compress, delta, smart_search, decision_trace | `mcp__contextstream__session(action="capture", event_type="decision", title="Use JWT", content="...")` |
| **`memory`** | action: create_event, get_event, update_event, delete_event, list_events, distill_event, create_node, get_node, update_node, delete_node, list_nodes, supersede_node, search, decisions, timeline, summary | `mcp__contextstream__memory(action="list_events", limit=10)` |
| **`graph`** | action: dependencies, impact, call_path, related, path, decisions, ingest, circular_dependencies, unused_code, contradictions | `mcp__contextstream__graph(action="impact", symbol_name="AuthService")` |
| **`project`** | action: list, get, create, update, index, overview, statistics, files, index_status, ingest_local | `mcp__contextstream__project(action="statistics")` |
| **`workspace`** | action: list, get, associate, bootstrap | `mcp__contextstream__workspace(action="list")` |
| **`reminder`** | action: list, active, create, snooze, complete, dismiss | `mcp__contextstream__reminder(action="active")` |
| **`integration`** | provider: slack/github/all; action: status, search, stats, activity, contributors, knowledge, summary, channels, discussions, sync_users, repos, issues | `mcp__contextstream__integration(provider="github", action="search", query="...")` |
| **`help`** | action: tools, auth, version, editor_rules, enable_bundle | `mcp__contextstream__help(action="tools")` |

---

### Why context_smart is Required (Even After session_init)

**Common mistake:** "session_init already gave me context, I don't need context_smart"

**This is WRONG. Here's why:**
- `session_init` returns the last ~10 items **BY TIME** (chronological)
- `context_smart` **SEARCHES** for items **RELEVANT to THIS message** (semantic)

**Example failure:**
- User asks: "how should I implement authentication?"
- Auth decisions were made 20 conversations ago
- `session_init` won't have it (too old, not in recent 10)
- `context_smart` FINDS it via semantic search

**Without context_smart, you WILL miss relevant older context.**

---

### Search & Code Intelligence (ContextStream-first)

⚠️ **STOP: Before using Search/Glob/Grep/Read/Explore** → Call `mcp__contextstream__search(mode="hybrid")` FIRST. Use local tools ONLY if ContextStream returns 0 results.

**❌ WRONG workflow (wastes tokens, slow):**

Grep "function" → Lire fichier1.ts → Lire fichier2.ts → Lire fichier3.ts → enfin comprendre


**✅ CORRECT workflow (fast, complete):**

mcp__contextstream__search(mode="hybrid", query="function implementation") → terminé (les résultats incluent le contexte)


**Why?** ContextStream search returns semantic matches + context + file locations in ONE call. Local tools require multiple round-trips.

**Search Mode Selection:**

| Need | Mode | Example |
|------|------|---------|
| Find code by meaning | `hybrid` | "authentication logic", "error handling" |
| Exact string/symbol | `keyword` | "UserAuthService", "API_KEY" |
| File patterns | `pattern` | "*.sql", "test_*.py" |
| ALL matches (grep-like) | `exhaustive` | "TODO", "FIXME" (find all occurrences) |
| Symbol renaming | `refactor` | "oldFunctionName" (word-boundary matching) |
| Conceptual search | `semantic` | "how does caching work" |

**Token Efficiency:** Use `output_format` to reduce response size:
- `full` (default): Full content for understanding code
- `paths`: File paths only (80% token savings) - use for file listings
- `minimal`: Compact format (60% savings) - use for refactoring
- `count`: Match counts only (90% savings) - use for quick checks

---

### Lessons (Past Mistakes)

- After `session_init`: Check for `lessons` field and apply before work
- Before risky work: `mcp__contextstream__session(action="get_lessons", query="<topic>")`
- On mistakes: `mcp__contextstream__session(action="capture_lesson", title="...", trigger="...", impact="...", prevention="...")`

---

### Plans & Tasks

When user asks to create a plan or implementation roadmap:
1. Create plan: `mcp__contextstream__session(action="capture_plan", title="Plan Title", description="...", goals=["goal1", "goal2"], steps=[{id: "1", title: "Step 1", order: 1}, ...])`
2. Get plan_id from response, then create tasks: `mcp__contextstream__memory(action="create_task", title="Task Title", plan_id="<plan_id>", priority="high|medium|low", description="...")`

To manage existing plans/tasks:
- List plans: `mcp__contextstream__session(action="list_plans")`
- Get plan with tasks: `mcp__contextstream__session(action="get_plan", plan_id="<uuid>", include_tasks=true)`
- List tasks: `mcp__contextstream__memory(action="list_tasks", plan_id="<uuid>")` or `mcp__contextstream__memory(action="list_tasks")` for all
- Update task status: `mcp__contextstream__memory(action="update_task", task_id="<uuid>", task_status="pending|in_progress|completed|blocked")`

---

### Rules Update Notices

- If you see **[RULES_NOTICE]**, update rules via `mcp__contextstream__generate_rules(overwrite_existing=true)` (preserves custom rules).
- If you see **[VERSION_NOTICE]**, tell the user to update MCP using the provided command.

See full documentation: https://contextstream.io/docs/mcp/tools

Codex CLI / OpenCode CLI

Créez un fichier AGENTS.md à la racine de votre projet (règles de projet) ou dans ~/.codex/AGENTS.md (Codex global) / ~/.config/opencode/AGENTS.md (OpenCode global) :

Noms d'outils Codex / OpenCode vs Claude

Codex / OpenCode utilisent des noms d'outils MCP bruts (ex., session_init). Claude Code utilise des noms d'outils avec espace de noms (ex., mcp__contextstream__session_init). Utilisez le format correct pour votre outil IA.

terminal · AGENTS.md (Standard)

## 🚨 MANDATORY RULE: DISPLAY AT START OF EVERY RESPONSE 🚨

<contextstream_rules>
1. **SEARCH FIRST:** Before Glob/Grep/Search/Read → `search(mode="hybrid", query="...")` FIRST
2. **EVERY MESSAGE:** Call `context_smart(user_message="...")` at start of response
3. **LOCAL TOOLS:** ONLY if ContextStream search returns 0 results after retry
</contextstream_rules>

**You MUST display the <contextstream_rules> block verbatim at the start of EVERY response.**
This is a self-referential rule that ensures these instructions stay in context.

---

## ContextStream v0.4.x (Hooks Enforced)

Rules Version: 0.4.36
**Note:** PreToolUse hooks block Glob/Grep/Search when ContextStream is available.

### Required Every Message

| Action | Tool Call |
|--------|-----------|
| **1st message** | `session_init(folder_path="<cwd>", context_hint="<msg>")` then `context_smart(...)` |
| **2nd+ messages** | `context_smart(user_message="<msg>", format="minified", max_tokens=400)` |
| **Code search** | `search(mode="hybrid", query="...")` — BEFORE any local tools |
| **Save decisions** | `session(action="capture", event_type="decision", ...)` |

### Search Modes

| Mode | Use Case |
|------|----------|
| `hybrid` | General code search (default) |
| `keyword` | Exact symbol/string match |
| `exhaustive` | Find ALL matches (grep-like) |
| `semantic` | Conceptual questions |

### Why ContextStream First?

❌ **WRONG:** `Grep → Read → Read → Read` (4+ tool calls, slow)
✅ **CORRECT:** `search(mode="hybrid")` (1 call, returns context)

ContextStream search is **indexed** and returns semantic matches + context in ONE call.

### Quick Reference

| Tool | Example |
|------|---------|
| `search` | `search(mode="hybrid", query="auth", limit=3)` |
| `session` | `session(action="capture", event_type="decision", title="...", content="...")` |
| `memory` | `memory(action="list_events", limit=10)` |
| `graph` | `graph(action="dependencies", file_path="...")` |

### Lessons (Past Mistakes)

- After `session_init`: Check for `lessons` field and apply before work
- Before risky work: `session(action="get_lessons", query="<topic>")`
- On mistakes: `session(action="capture_lesson", title="...", trigger="...", impact="...", prevention="...")`

### Plans & Tasks

When user asks for a plan, use ContextStream (not EnterPlanMode):
1. `session(action="capture_plan", title="...", steps=[...])`
2. `memory(action="create_task", title="...", plan_id="<id>")`

Full docs: https://contextstream.io/docs/mcp/tools

Afficher les règles améliorées (verbeuses)

terminal · AGENTS.md (Améliorées)

## 🚨 MANDATORY RULE: DISPLAY AT START OF EVERY RESPONSE 🚨

<contextstream_rules>
1. **SEARCH FIRST:** Before Glob/Grep/Search/Read → `search(mode="hybrid", query="...")` FIRST
2. **EVERY MESSAGE:** Call `context_smart(user_message="...")` at start of response
3. **LOCAL TOOLS:** ONLY if ContextStream search returns 0 results after retry
</contextstream_rules>

**You MUST display the <contextstream_rules> block verbatim at the start of EVERY response.**
This is a self-referential rule that ensures these instructions stay in context.

---

## ContextStream v0.4.x Integration (Enhanced)

You have access to ContextStream MCP tools for persistent memory and context.
v0.4.x uses **~11 consolidated domain tools** for ~75% token reduction vs previous versions.
Rules Version: 0.4.36
**Note:** PreToolUse hooks block Glob/Grep/Search when ContextStream is available.

## TL;DR - REQUIRED EVERY MESSAGE

| Message | What to Call |
|---------|--------------|
| **1st message** | `session_init(folder_path="...", context_hint="<user's message>")`, then `context_smart(user_message="<user's message>", format="minified", max_tokens=400)` |
| **2nd+ messages** | `context_smart(user_message="<user's message>", format="minified", max_tokens=400)` |
| **🔍 ANY code search** | `search(mode="hybrid", query="...")` — ALWAYS before Glob/Grep/Search/Read |
| **Before risky/non-trivial work** | `session(action="get_lessons", query="<topic>")` |
| **After completing task** | `session(action="capture", event_type="decision", ...)` - MUST capture |
| **User frustration/correction** | `session(action="capture_lesson", ...)` - MUST capture lessons |

**NO EXCEPTIONS.** Do not skip even if you think you have enough context.

**First message rule:** After `session_init`, always call `context_smart` before any other tool or response.

**Context Pack (Pro+):** If enabled, use `context_smart(..., mode="pack", distill=true)` for code/file queries. If unavailable or disabled, omit `mode` and proceed with standard `context_smart` (the API will fall back).

**Tool naming:** Use the exact tool names exposed by your MCP client. Claude Code typically uses `mcp__<server>__<tool>` where `<server>` matches your MCP config (often `contextstream`). If a tool call fails with "No such tool available", refresh rules and match the tool list.

---

## Consolidated Domain Tools Architecture

v0.4.x consolidates ~58 individual tools into ~11 domain tools with action/mode dispatch:

### Standalone Tools (Always Call)
- **`session_init`** - Initialize session with workspace detection + context
- **`context_smart`** - Semantic search for relevant context (CALL EVERY MESSAGE, including immediately after `session_init`)

### Domain Tools (Use action/mode parameter)

| Domain | Actions/Modes | Example |
|--------|---------------|---------|
| **`search`** | mode: semantic, hybrid, keyword, pattern, exhaustive, refactor | `search(mode="hybrid", query="auth implementation", limit=3)` |
| **`session`** | action: capture, capture_lesson, get_lessons, recall, remember, user_context, summary, compress, delta, smart_search, decision_trace | `session(action="capture", event_type="decision", title="Use JWT", content="...")` |
| **`memory`** | action: create_event, get_event, update_event, delete_event, list_events, distill_event, create_node, get_node, update_node, delete_node, list_nodes, supersede_node, search, decisions, timeline, summary | `memory(action="list_events", limit=10)` |
| **`graph`** | action: dependencies, impact, call_path, related, path, decisions, ingest, circular_dependencies, unused_code, contradictions | `graph(action="impact", symbol_name="AuthService")` |
| **`project`** | action: list, get, create, update, index, overview, statistics, files, index_status, ingest_local | `project(action="statistics")` |
| **`workspace`** | action: list, get, associate, bootstrap | `workspace(action="list")` |
| **`reminder`** | action: list, active, create, snooze, complete, dismiss | `reminder(action="active")` |
| **`integration`** | provider: slack/github/all; action: status, search, stats, activity, contributors, knowledge, summary, channels, discussions, sync_users, repos, issues | `integration(provider="github", action="search", query="...")` |
| **`help`** | action: tools, auth, version, editor_rules, enable_bundle | `help(action="tools")` |

---

### Why context_smart is Required (Even After session_init)

**Common mistake:** "session_init already gave me context, I don't need context_smart"

**This is WRONG. Here's why:**
- `session_init` returns the last ~10 items **BY TIME** (chronological)
- `context_smart` **SEARCHES** for items **RELEVANT to THIS message** (semantic)

**Example failure:**
- User asks: "how should I implement authentication?"
- Auth decisions were made 20 conversations ago
- `session_init` won't have it (too old, not in recent 10)
- `context_smart` FINDS it via semantic search

**Without context_smart, you WILL miss relevant older context.**

---

### Search & Code Intelligence (ContextStream-first)

⚠️ **STOP: Before using Search/Glob/Grep/Read/Explore** → Call `search(mode="hybrid")` FIRST. Use local tools ONLY if ContextStream returns 0 results.

**❌ WRONG workflow (wastes tokens, slow):**

Grep "function" → Lire fichier1.ts → Lire fichier2.ts → Lire fichier3.ts → enfin comprendre


**✅ CORRECT workflow (fast, complete):**

search(mode="hybrid", query="function implementation") → terminé (les résultats incluent le contexte)


**Why?** ContextStream search returns semantic matches + context + file locations in ONE call. Local tools require multiple round-trips.

**Search Mode Selection:**

| Need | Mode | Example |
|------|------|---------|
| Find code by meaning | `hybrid` | "authentication logic", "error handling" |
| Exact string/symbol | `keyword` | "UserAuthService", "API_KEY" |
| File patterns | `pattern` | "*.sql", "test_*.py" |
| ALL matches (grep-like) | `exhaustive` | "TODO", "FIXME" (find all occurrences) |
| Symbol renaming | `refactor` | "oldFunctionName" (word-boundary matching) |
| Conceptual search | `semantic` | "how does caching work" |

**Token Efficiency:** Use `output_format` to reduce response size:
- `full` (default): Full content for understanding code
- `paths`: File paths only (80% token savings) - use for file listings
- `minimal`: Compact format (60% savings) - use for refactoring
- `count`: Match counts only (90% savings) - use for quick checks

---

### Lessons (Past Mistakes)

- After `session_init`: Check for `lessons` field and apply before work
- Before risky work: `session(action="get_lessons", query="<topic>")`
- On mistakes: `session(action="capture_lesson", title="...", trigger="...", impact="...", prevention="...")`

---

### Plans & Tasks

When user asks to create a plan or implementation roadmap:
1. Create plan: `session(action="capture_plan", title="Plan Title", description="...", goals=["goal1", "goal2"], steps=[{id: "1", title: "Step 1", order: 1}, ...])`
2. Get plan_id from response, then create tasks: `memory(action="create_task", title="Task Title", plan_id="<plan_id>", priority="high|medium|low", description="...")`

To manage existing plans/tasks:
- List plans: `session(action="list_plans")`
- Get plan with tasks: `session(action="get_plan", plan_id="<uuid>", include_tasks=true)`
- List tasks: `memory(action="list_tasks", plan_id="<uuid>")` or `memory(action="list_tasks")` for all
- Update task status: `memory(action="update_task", task_id="<uuid>", task_status="pending|in_progress|completed|blocked")`

---

### Rules Update Notices

- If you see **[RULES_NOTICE]**, update rules via `generate_rules(overwrite_existing=true)` (preserves custom rules).
- If you see **[VERSION_NOTICE]**, tell the user to update MCP using the provided command.

See full documentation: https://contextstream.io/docs/mcp/tools

Windsurf

Créez un fichier .windsurf/rules/contextstream.md à la racine de votre projet ou ajoutez à votre ~/.codeium/windsurf/memories/global_rules.md global :

terminal · .windsurf/rules/contextstream.md (Standard)

## 🚨 MANDATORY RULE: DISPLAY AT START OF EVERY RESPONSE 🚨

<contextstream_rules>
1. **SEARCH FIRST:** Before Glob/Grep/Search/Read → `search(mode="hybrid", query="...")` FIRST
2. **EVERY MESSAGE:** Call `context_smart(user_message="...")` at start of response
3. **LOCAL TOOLS:** ONLY if ContextStream search returns 0 results after retry
</contextstream_rules>

**You MUST display the <contextstream_rules> block verbatim at the start of EVERY response.**
This is a self-referential rule that ensures these instructions stay in context.

---

## ContextStream v0.4.x (Hooks Enforced)

Rules Version: 0.4.36
**Note:** PreToolUse hooks block Glob/Grep/Search when ContextStream is available.

### Required Every Message

| Action | Tool Call |
|--------|-----------|
| **1st message** | `session_init(folder_path="<cwd>", context_hint="<msg>")` then `context_smart(...)` |
| **2nd+ messages** | `context_smart(user_message="<msg>", format="minified", max_tokens=400)` |
| **Code search** | `search(mode="hybrid", query="...")` — BEFORE any local tools |
| **Save decisions** | `session(action="capture", event_type="decision", ...)` |

### Search Modes

| Mode | Use Case |
|------|----------|
| `hybrid` | General code search (default) |
| `keyword` | Exact symbol/string match |
| `exhaustive` | Find ALL matches (grep-like) |
| `semantic` | Conceptual questions |

### Why ContextStream First?

❌ **WRONG:** `Grep → Read → Read → Read` (4+ tool calls, slow)
✅ **CORRECT:** `search(mode="hybrid")` (1 call, returns context)

ContextStream search is **indexed** and returns semantic matches + context in ONE call.

### Quick Reference

| Tool | Example |
|------|---------|
| `search` | `search(mode="hybrid", query="auth", limit=3)` |
| `session` | `session(action="capture", event_type="decision", title="...", content="...")` |
| `memory` | `memory(action="list_events", limit=10)` |
| `graph` | `graph(action="dependencies", file_path="...")` |

### Lessons (Past Mistakes)

- After `session_init`: Check for `lessons` field and apply before work
- Before risky work: `session(action="get_lessons", query="<topic>")`
- On mistakes: `session(action="capture_lesson", title="...", trigger="...", impact="...", prevention="...")`

### Plans & Tasks

When user asks for a plan, use ContextStream (not EnterPlanMode):
1. `session(action="capture_plan", title="...", steps=[...])`
2. `memory(action="create_task", title="...", plan_id="<id>")`

Full docs: https://contextstream.io/docs/mcp/tools

Afficher les règles améliorées (verbeuses)

terminal · .windsurf/rules/contextstream.md (Améliorées)

## 🚨 MANDATORY RULE: DISPLAY AT START OF EVERY RESPONSE 🚨

<contextstream_rules>
1. **SEARCH FIRST:** Before Glob/Grep/Search/Read → `search(mode="hybrid", query="...")` FIRST
2. **EVERY MESSAGE:** Call `context_smart(user_message="...")` at start of response
3. **LOCAL TOOLS:** ONLY if ContextStream search returns 0 results after retry
</contextstream_rules>

**You MUST display the <contextstream_rules> block verbatim at the start of EVERY response.**
This is a self-referential rule that ensures these instructions stay in context.

---

## ContextStream v0.4.x Integration (Enhanced)

You have access to ContextStream MCP tools for persistent memory and context.
v0.4.x uses **~11 consolidated domain tools** for ~75% token reduction vs previous versions.
Rules Version: 0.4.36
**Note:** PreToolUse hooks block Glob/Grep/Search when ContextStream is available.

## TL;DR - REQUIRED EVERY MESSAGE

| Message | What to Call |
|---------|--------------|
| **1st message** | `session_init(folder_path="...", context_hint="<user's message>")`, then `context_smart(user_message="<user's message>", format="minified", max_tokens=400)` |
| **2nd+ messages** | `context_smart(user_message="<user's message>", format="minified", max_tokens=400)` |
| **🔍 ANY code search** | `search(mode="hybrid", query="...")` — ALWAYS before Glob/Grep/Search/Read |
| **Before risky/non-trivial work** | `session(action="get_lessons", query="<topic>")` |
| **After completing task** | `session(action="capture", event_type="decision", ...)` - MUST capture |
| **User frustration/correction** | `session(action="capture_lesson", ...)` - MUST capture lessons |

**NO EXCEPTIONS.** Do not skip even if you think you have enough context.

**First message rule:** After `session_init`, always call `context_smart` before any other tool or response.

**Context Pack (Pro+):** If enabled, use `context_smart(..., mode="pack", distill=true)` for code/file queries. If unavailable or disabled, omit `mode` and proceed with standard `context_smart` (the API will fall back).

**Tool naming:** Use the exact tool names exposed by your MCP client. Claude Code typically uses `mcp__<server>__<tool>` where `<server>` matches your MCP config (often `contextstream`). If a tool call fails with "No such tool available", refresh rules and match the tool list.

---

## Consolidated Domain Tools Architecture

v0.4.x consolidates ~58 individual tools into ~11 domain tools with action/mode dispatch:

### Standalone Tools (Always Call)
- **`session_init`** - Initialize session with workspace detection + context
- **`context_smart`** - Semantic search for relevant context (CALL EVERY MESSAGE, including immediately after `session_init`)

### Domain Tools (Use action/mode parameter)

| Domain | Actions/Modes | Example |
|--------|---------------|---------|
| **`search`** | mode: semantic, hybrid, keyword, pattern, exhaustive, refactor | `search(mode="hybrid", query="auth implementation", limit=3)` |
| **`session`** | action: capture, capture_lesson, get_lessons, recall, remember, user_context, summary, compress, delta, smart_search, decision_trace | `session(action="capture", event_type="decision", title="Use JWT", content="...")` |
| **`memory`** | action: create_event, get_event, update_event, delete_event, list_events, distill_event, create_node, get_node, update_node, delete_node, list_nodes, supersede_node, search, decisions, timeline, summary | `memory(action="list_events", limit=10)` |
| **`graph`** | action: dependencies, impact, call_path, related, path, decisions, ingest, circular_dependencies, unused_code, contradictions | `graph(action="impact", symbol_name="AuthService")` |
| **`project`** | action: list, get, create, update, index, overview, statistics, files, index_status, ingest_local | `project(action="statistics")` |
| **`workspace`** | action: list, get, associate, bootstrap | `workspace(action="list")` |
| **`reminder`** | action: list, active, create, snooze, complete, dismiss | `reminder(action="active")` |
| **`integration`** | provider: slack/github/all; action: status, search, stats, activity, contributors, knowledge, summary, channels, discussions, sync_users, repos, issues | `integration(provider="github", action="search", query="...")` |
| **`help`** | action: tools, auth, version, editor_rules, enable_bundle | `help(action="tools")` |

---

### Why context_smart is Required (Even After session_init)

**Common mistake:** "session_init already gave me context, I don't need context_smart"

**This is WRONG. Here's why:**
- `session_init` returns the last ~10 items **BY TIME** (chronological)
- `context_smart` **SEARCHES** for items **RELEVANT to THIS message** (semantic)

**Example failure:**
- User asks: "how should I implement authentication?"
- Auth decisions were made 20 conversations ago
- `session_init` won't have it (too old, not in recent 10)
- `context_smart` FINDS it via semantic search

**Without context_smart, you WILL miss relevant older context.**

---

### Search & Code Intelligence (ContextStream-first)

⚠️ **STOP: Before using Search/Glob/Grep/Read/Explore** → Call `search(mode="hybrid")` FIRST. Use local tools ONLY if ContextStream returns 0 results.

**❌ WRONG workflow (wastes tokens, slow):**

Grep "function" → Lire fichier1.ts → Lire fichier2.ts → Lire fichier3.ts → enfin comprendre


**✅ CORRECT workflow (fast, complete):**

search(mode="hybrid", query="function implementation") → terminé (les résultats incluent le contexte)


**Why?** ContextStream search returns semantic matches + context + file locations in ONE call. Local tools require multiple round-trips.

**Search Mode Selection:**

| Need | Mode | Example |
|------|------|---------|
| Find code by meaning | `hybrid` | "authentication logic", "error handling" |
| Exact string/symbol | `keyword` | "UserAuthService", "API_KEY" |
| File patterns | `pattern` | "*.sql", "test_*.py" |
| ALL matches (grep-like) | `exhaustive` | "TODO", "FIXME" (find all occurrences) |
| Symbol renaming | `refactor` | "oldFunctionName" (word-boundary matching) |
| Conceptual search | `semantic` | "how does caching work" |

**Token Efficiency:** Use `output_format` to reduce response size:
- `full` (default): Full content for understanding code
- `paths`: File paths only (80% token savings) - use for file listings
- `minimal`: Compact format (60% savings) - use for refactoring
- `count`: Match counts only (90% savings) - use for quick checks

---

### Lessons (Past Mistakes)

- After `session_init`: Check for `lessons` field and apply before work
- Before risky work: `session(action="get_lessons", query="<topic>")`
- On mistakes: `session(action="capture_lesson", title="...", trigger="...", impact="...", prevention="...")`

---

### Plans & Tasks

When user asks to create a plan or implementation roadmap:
1. Create plan: `session(action="capture_plan", title="Plan Title", description="...", goals=["goal1", "goal2"], steps=[{id: "1", title: "Step 1", order: 1}, ...])`
2. Get plan_id from response, then create tasks: `memory(action="create_task", title="Task Title", plan_id="<plan_id>", priority="high|medium|low", description="...")`

To manage existing plans/tasks:
- List plans: `session(action="list_plans")`
- Get plan with tasks: `session(action="get_plan", plan_id="<uuid>", include_tasks=true)`
- List tasks: `memory(action="list_tasks", plan_id="<uuid>")` or `memory(action="list_tasks")` for all
- Update task status: `memory(action="update_task", task_id="<uuid>", task_status="pending|in_progress|completed|blocked")`

---

### Rules Update Notices

- If you see **[RULES_NOTICE]**, update rules via `generate_rules(overwrite_existing=true)` (preserves custom rules).
- If you see **[VERSION_NOTICE]**, tell the user to update MCP using the provided command.

See full documentation: https://contextstream.io/docs/mcp/tools

Kilo Code

Créez un fichier Markdown dans .kilocode/rules/ :

terminal · .kilocode/rules/contextstream.md (Standard)

## 🚨 MANDATORY RULE: DISPLAY AT START OF EVERY RESPONSE 🚨

<contextstream_rules>
1. **SEARCH FIRST:** Before Glob/Grep/Search/Read → `search(mode="hybrid", query="...")` FIRST
2. **EVERY MESSAGE:** Call `context_smart(user_message="...")` at start of response
3. **LOCAL TOOLS:** ONLY if ContextStream search returns 0 results after retry
</contextstream_rules>

**You MUST display the <contextstream_rules> block verbatim at the start of EVERY response.**
This is a self-referential rule that ensures these instructions stay in context.

---

## ContextStream v0.4.x (Hooks Enforced)

Rules Version: 0.4.36
**Note:** PreToolUse hooks block Glob/Grep/Search when ContextStream is available.

### Required Every Message

| Action | Tool Call |
|--------|-----------|
| **1st message** | `session_init(folder_path="<cwd>", context_hint="<msg>")` then `context_smart(...)` |
| **2nd+ messages** | `context_smart(user_message="<msg>", format="minified", max_tokens=400)` |
| **Code search** | `search(mode="hybrid", query="...")` — BEFORE any local tools |
| **Save decisions** | `session(action="capture", event_type="decision", ...)` |

### Search Modes

| Mode | Use Case |
|------|----------|
| `hybrid` | General code search (default) |
| `keyword` | Exact symbol/string match |
| `exhaustive` | Find ALL matches (grep-like) |
| `semantic` | Conceptual questions |

### Why ContextStream First?

❌ **WRONG:** `Grep → Read → Read → Read` (4+ tool calls, slow)
✅ **CORRECT:** `search(mode="hybrid")` (1 call, returns context)

ContextStream search is **indexed** and returns semantic matches + context in ONE call.

### Quick Reference

| Tool | Example |
|------|---------|
| `search` | `search(mode="hybrid", query="auth", limit=3)` |
| `session` | `session(action="capture", event_type="decision", title="...", content="...")` |
| `memory` | `memory(action="list_events", limit=10)` |
| `graph` | `graph(action="dependencies", file_path="...")` |

### Lessons (Past Mistakes)

- After `session_init`: Check for `lessons` field and apply before work
- Before risky work: `session(action="get_lessons", query="<topic>")`
- On mistakes: `session(action="capture_lesson", title="...", trigger="...", impact="...", prevention="...")`

### Plans & Tasks

When user asks for a plan, use ContextStream (not EnterPlanMode):
1. `session(action="capture_plan", title="...", steps=[...])`
2. `memory(action="create_task", title="...", plan_id="<id>")`

Full docs: https://contextstream.io/docs/mcp/tools

Afficher les règles améliorées (verbeuses)

terminal · .kilocode/rules/contextstream.md (Améliorées)

## 🚨 MANDATORY RULE: DISPLAY AT START OF EVERY RESPONSE 🚨

<contextstream_rules>
1. **SEARCH FIRST:** Before Glob/Grep/Search/Read → `search(mode="hybrid", query="...")` FIRST
2. **EVERY MESSAGE:** Call `context_smart(user_message="...")` at start of response
3. **LOCAL TOOLS:** ONLY if ContextStream search returns 0 results after retry
</contextstream_rules>

**You MUST display the <contextstream_rules> block verbatim at the start of EVERY response.**
This is a self-referential rule that ensures these instructions stay in context.

---

## ContextStream v0.4.x Integration (Enhanced)

You have access to ContextStream MCP tools for persistent memory and context.
v0.4.x uses **~11 consolidated domain tools** for ~75% token reduction vs previous versions.
Rules Version: 0.4.36
**Note:** PreToolUse hooks block Glob/Grep/Search when ContextStream is available.

## TL;DR - REQUIRED EVERY MESSAGE

| Message | What to Call |
|---------|--------------|
| **1st message** | `session_init(folder_path="...", context_hint="<user's message>")`, then `context_smart(user_message="<user's message>", format="minified", max_tokens=400)` |
| **2nd+ messages** | `context_smart(user_message="<user's message>", format="minified", max_tokens=400)` |
| **🔍 ANY code search** | `search(mode="hybrid", query="...")` — ALWAYS before Glob/Grep/Search/Read |
| **Before risky/non-trivial work** | `session(action="get_lessons", query="<topic>")` |
| **After completing task** | `session(action="capture", event_type="decision", ...)` - MUST capture |
| **User frustration/correction** | `session(action="capture_lesson", ...)` - MUST capture lessons |

**NO EXCEPTIONS.** Do not skip even if you think you have enough context.

**First message rule:** After `session_init`, always call `context_smart` before any other tool or response.

**Context Pack (Pro+):** If enabled, use `context_smart(..., mode="pack", distill=true)` for code/file queries. If unavailable or disabled, omit `mode` and proceed with standard `context_smart` (the API will fall back).

**Tool naming:** Use the exact tool names exposed by your MCP client. Claude Code typically uses `mcp__<server>__<tool>` where `<server>` matches your MCP config (often `contextstream`). If a tool call fails with "No such tool available", refresh rules and match the tool list.

---

## Consolidated Domain Tools Architecture

v0.4.x consolidates ~58 individual tools into ~11 domain tools with action/mode dispatch:

### Standalone Tools (Always Call)
- **`session_init`** - Initialize session with workspace detection + context
- **`context_smart`** - Semantic search for relevant context (CALL EVERY MESSAGE, including immediately after `session_init`)

### Domain Tools (Use action/mode parameter)

| Domain | Actions/Modes | Example |
|--------|---------------|---------|
| **`search`** | mode: semantic, hybrid, keyword, pattern, exhaustive, refactor | `search(mode="hybrid", query="auth implementation", limit=3)` |
| **`session`** | action: capture, capture_lesson, get_lessons, recall, remember, user_context, summary, compress, delta, smart_search, decision_trace | `session(action="capture", event_type="decision", title="Use JWT", content="...")` |
| **`memory`** | action: create_event, get_event, update_event, delete_event, list_events, distill_event, create_node, get_node, update_node, delete_node, list_nodes, supersede_node, search, decisions, timeline, summary | `memory(action="list_events", limit=10)` |
| **`graph`** | action: dependencies, impact, call_path, related, path, decisions, ingest, circular_dependencies, unused_code, contradictions | `graph(action="impact", symbol_name="AuthService")` |
| **`project`** | action: list, get, create, update, index, overview, statistics, files, index_status, ingest_local | `project(action="statistics")` |
| **`workspace`** | action: list, get, associate, bootstrap | `workspace(action="list")` |
| **`reminder`** | action: list, active, create, snooze, complete, dismiss | `reminder(action="active")` |
| **`integration`** | provider: slack/github/all; action: status, search, stats, activity, contributors, knowledge, summary, channels, discussions, sync_users, repos, issues | `integration(provider="github", action="search", query="...")` |
| **`help`** | action: tools, auth, version, editor_rules, enable_bundle | `help(action="tools")` |

---

### Why context_smart is Required (Even After session_init)

**Common mistake:** "session_init already gave me context, I don't need context_smart"

**This is WRONG. Here's why:**
- `session_init` returns the last ~10 items **BY TIME** (chronological)
- `context_smart` **SEARCHES** for items **RELEVANT to THIS message** (semantic)

**Example failure:**
- User asks: "how should I implement authentication?"
- Auth decisions were made 20 conversations ago
- `session_init` won't have it (too old, not in recent 10)
- `context_smart` FINDS it via semantic search

**Without context_smart, you WILL miss relevant older context.**

---

### Search & Code Intelligence (ContextStream-first)

⚠️ **STOP: Before using Search/Glob/Grep/Read/Explore** → Call `search(mode="hybrid")` FIRST. Use local tools ONLY if ContextStream returns 0 results.

**❌ WRONG workflow (wastes tokens, slow):**

Grep "function" → Lire fichier1.ts → Lire fichier2.ts → Lire fichier3.ts → enfin comprendre


**✅ CORRECT workflow (fast, complete):**

search(mode="hybrid", query="function implementation") → terminé (les résultats incluent le contexte)


**Why?** ContextStream search returns semantic matches + context + file locations in ONE call. Local tools require multiple round-trips.

**Search Mode Selection:**

| Need | Mode | Example |
|------|------|---------|
| Find code by meaning | `hybrid` | "authentication logic", "error handling" |
| Exact string/symbol | `keyword` | "UserAuthService", "API_KEY" |
| File patterns | `pattern` | "*.sql", "test_*.py" |
| ALL matches (grep-like) | `exhaustive` | "TODO", "FIXME" (find all occurrences) |
| Symbol renaming | `refactor` | "oldFunctionName" (word-boundary matching) |
| Conceptual search | `semantic` | "how does caching work" |

**Token Efficiency:** Use `output_format` to reduce response size:
- `full` (default): Full content for understanding code
- `paths`: File paths only (80% token savings) - use for file listings
- `minimal`: Compact format (60% savings) - use for refactoring
- `count`: Match counts only (90% savings) - use for quick checks

---

### Lessons (Past Mistakes)

- After `session_init`: Check for `lessons` field and apply before work
- Before risky work: `session(action="get_lessons", query="<topic>")`
- On mistakes: `session(action="capture_lesson", title="...", trigger="...", impact="...", prevention="...")`

---

### Plans & Tasks

When user asks to create a plan or implementation roadmap:
1. Create plan: `session(action="capture_plan", title="Plan Title", description="...", goals=["goal1", "goal2"], steps=[{id: "1", title: "Step 1", order: 1}, ...])`
2. Get plan_id from response, then create tasks: `memory(action="create_task", title="Task Title", plan_id="<plan_id>", priority="high|medium|low", description="...")`

To manage existing plans/tasks:
- List plans: `session(action="list_plans")`
- Get plan with tasks: `session(action="get_plan", plan_id="<uuid>", include_tasks=true)`
- List tasks: `memory(action="list_tasks", plan_id="<uuid>")` or `memory(action="list_tasks")` for all
- Update task status: `memory(action="update_task", task_id="<uuid>", task_status="pending|in_progress|completed|blocked")`

---

### Rules Update Notices

- If you see **[RULES_NOTICE]**, update rules via `generate_rules(overwrite_existing=true)` (preserves custom rules).
- If you see **[VERSION_NOTICE]**, tell the user to update MCP using the provided command.

See full documentation: https://contextstream.io/docs/mcp/tools

Cline

Créez un fichier .clinerules à la racine de votre projet ou utilisez le dossier .clinerules/ :

terminal · .clinerules (Standard)

## 🚨 MANDATORY RULE: DISPLAY AT START OF EVERY RESPONSE 🚨

<contextstream_rules>
1. **SEARCH FIRST:** Before Glob/Grep/Search/Read → `search(mode="hybrid", query="...")` FIRST
2. **EVERY MESSAGE:** Call `context_smart(user_message="...")` at start of response
3. **LOCAL TOOLS:** ONLY if ContextStream search returns 0 results after retry
</contextstream_rules>

**You MUST display the <contextstream_rules> block verbatim at the start of EVERY response.**
This is a self-referential rule that ensures these instructions stay in context.

---

## ContextStream v0.4.x (Hooks Enforced)

Rules Version: 0.4.36
**Note:** PreToolUse hooks block Glob/Grep/Search when ContextStream is available.

### Required Every Message

| Action | Tool Call |
|--------|-----------|
| **1st message** | `session_init(folder_path="<cwd>", context_hint="<msg>")` then `context_smart(...)` |
| **2nd+ messages** | `context_smart(user_message="<msg>", format="minified", max_tokens=400)` |
| **Code search** | `search(mode="hybrid", query="...")` — BEFORE any local tools |
| **Save decisions** | `session(action="capture", event_type="decision", ...)` |

### Search Modes

| Mode | Use Case |
|------|----------|
| `hybrid` | General code search (default) |
| `keyword` | Exact symbol/string match |
| `exhaustive` | Find ALL matches (grep-like) |
| `semantic` | Conceptual questions |

### Why ContextStream First?

❌ **WRONG:** `Grep → Read → Read → Read` (4+ tool calls, slow)
✅ **CORRECT:** `search(mode="hybrid")` (1 call, returns context)

ContextStream search is **indexed** and returns semantic matches + context in ONE call.

### Quick Reference

| Tool | Example |
|------|---------|
| `search` | `search(mode="hybrid", query="auth", limit=3)` |
| `session` | `session(action="capture", event_type="decision", title="...", content="...")` |
| `memory` | `memory(action="list_events", limit=10)` |
| `graph` | `graph(action="dependencies", file_path="...")` |

### Lessons (Past Mistakes)

- After `session_init`: Check for `lessons` field and apply before work
- Before risky work: `session(action="get_lessons", query="<topic>")`
- On mistakes: `session(action="capture_lesson", title="...", trigger="...", impact="...", prevention="...")`

### Plans & Tasks

When user asks for a plan, use ContextStream (not EnterPlanMode):
1. `session(action="capture_plan", title="...", steps=[...])`
2. `memory(action="create_task", title="...", plan_id="<id>")`

Full docs: https://contextstream.io/docs/mcp/tools

Afficher les règles améliorées (verbeuses)

terminal · .clinerules (Améliorées)

## 🚨 MANDATORY RULE: DISPLAY AT START OF EVERY RESPONSE 🚨

<contextstream_rules>
1. **SEARCH FIRST:** Before Glob/Grep/Search/Read → `search(mode="hybrid", query="...")` FIRST
2. **EVERY MESSAGE:** Call `context_smart(user_message="...")` at start of response
3. **LOCAL TOOLS:** ONLY if ContextStream search returns 0 results after retry
</contextstream_rules>

**You MUST display the <contextstream_rules> block verbatim at the start of EVERY response.**
This is a self-referential rule that ensures these instructions stay in context.

---

## ContextStream v0.4.x Integration (Enhanced)

You have access to ContextStream MCP tools for persistent memory and context.
v0.4.x uses **~11 consolidated domain tools** for ~75% token reduction vs previous versions.
Rules Version: 0.4.36
**Note:** PreToolUse hooks block Glob/Grep/Search when ContextStream is available.

## TL;DR - REQUIRED EVERY MESSAGE

| Message | What to Call |
|---------|--------------|
| **1st message** | `session_init(folder_path="...", context_hint="<user's message>")`, then `context_smart(user_message="<user's message>", format="minified", max_tokens=400)` |
| **2nd+ messages** | `context_smart(user_message="<user's message>", format="minified", max_tokens=400)` |
| **🔍 ANY code search** | `search(mode="hybrid", query="...")` — ALWAYS before Glob/Grep/Search/Read |
| **Before risky/non-trivial work** | `session(action="get_lessons", query="<topic>")` |
| **After completing task** | `session(action="capture", event_type="decision", ...)` - MUST capture |
| **User frustration/correction** | `session(action="capture_lesson", ...)` - MUST capture lessons |

**NO EXCEPTIONS.** Do not skip even if you think you have enough context.

**First message rule:** After `session_init`, always call `context_smart` before any other tool or response.

**Context Pack (Pro+):** If enabled, use `context_smart(..., mode="pack", distill=true)` for code/file queries. If unavailable or disabled, omit `mode` and proceed with standard `context_smart` (the API will fall back).

**Tool naming:** Use the exact tool names exposed by your MCP client. Claude Code typically uses `mcp__<server>__<tool>` where `<server>` matches your MCP config (often `contextstream`). If a tool call fails with "No such tool available", refresh rules and match the tool list.

---

## Consolidated Domain Tools Architecture

v0.4.x consolidates ~58 individual tools into ~11 domain tools with action/mode dispatch:

### Standalone Tools (Always Call)
- **`session_init`** - Initialize session with workspace detection + context
- **`context_smart`** - Semantic search for relevant context (CALL EVERY MESSAGE, including immediately after `session_init`)

### Domain Tools (Use action/mode parameter)

| Domain | Actions/Modes | Example |
|--------|---------------|---------|
| **`search`** | mode: semantic, hybrid, keyword, pattern, exhaustive, refactor | `search(mode="hybrid", query="auth implementation", limit=3)` |
| **`session`** | action: capture, capture_lesson, get_lessons, recall, remember, user_context, summary, compress, delta, smart_search, decision_trace | `session(action="capture", event_type="decision", title="Use JWT", content="...")` |
| **`memory`** | action: create_event, get_event, update_event, delete_event, list_events, distill_event, create_node, get_node, update_node, delete_node, list_nodes, supersede_node, search, decisions, timeline, summary | `memory(action="list_events", limit=10)` |
| **`graph`** | action: dependencies, impact, call_path, related, path, decisions, ingest, circular_dependencies, unused_code, contradictions | `graph(action="impact", symbol_name="AuthService")` |
| **`project`** | action: list, get, create, update, index, overview, statistics, files, index_status, ingest_local | `project(action="statistics")` |
| **`workspace`** | action: list, get, associate, bootstrap | `workspace(action="list")` |
| **`reminder`** | action: list, active, create, snooze, complete, dismiss | `reminder(action="active")` |
| **`integration`** | provider: slack/github/all; action: status, search, stats, activity, contributors, knowledge, summary, channels, discussions, sync_users, repos, issues | `integration(provider="github", action="search", query="...")` |
| **`help`** | action: tools, auth, version, editor_rules, enable_bundle | `help(action="tools")` |

---

### Why context_smart is Required (Even After session_init)

**Common mistake:** "session_init already gave me context, I don't need context_smart"

**This is WRONG. Here's why:**
- `session_init` returns the last ~10 items **BY TIME** (chronological)
- `context_smart` **SEARCHES** for items **RELEVANT to THIS message** (semantic)

**Example failure:**
- User asks: "how should I implement authentication?"
- Auth decisions were made 20 conversations ago
- `session_init` won't have it (too old, not in recent 10)
- `context_smart` FINDS it via semantic search

**Without context_smart, you WILL miss relevant older context.**

---

### Search & Code Intelligence (ContextStream-first)

⚠️ **STOP: Before using Search/Glob/Grep/Read/Explore** → Call `search(mode="hybrid")` FIRST. Use local tools ONLY if ContextStream returns 0 results.

**❌ WRONG workflow (wastes tokens, slow):**

Grep "function" → Lire fichier1.ts → Lire fichier2.ts → Lire fichier3.ts → enfin comprendre


**✅ CORRECT workflow (fast, complete):**

search(mode="hybrid", query="function implementation") → terminé (les résultats incluent le contexte)


**Why?** ContextStream search returns semantic matches + context + file locations in ONE call. Local tools require multiple round-trips.

**Search Mode Selection:**

| Need | Mode | Example |
|------|------|---------|
| Find code by meaning | `hybrid` | "authentication logic", "error handling" |
| Exact string/symbol | `keyword` | "UserAuthService", "API_KEY" |
| File patterns | `pattern` | "*.sql", "test_*.py" |
| ALL matches (grep-like) | `exhaustive` | "TODO", "FIXME" (find all occurrences) |
| Symbol renaming | `refactor` | "oldFunctionName" (word-boundary matching) |
| Conceptual search | `semantic` | "how does caching work" |

**Token Efficiency:** Use `output_format` to reduce response size:
- `full` (default): Full content for understanding code
- `paths`: File paths only (80% token savings) - use for file listings
- `minimal`: Compact format (60% savings) - use for refactoring
- `count`: Match counts only (90% savings) - use for quick checks

---

### Lessons (Past Mistakes)

- After `session_init`: Check for `lessons` field and apply before work
- Before risky work: `session(action="get_lessons", query="<topic>")`
- On mistakes: `session(action="capture_lesson", title="...", trigger="...", impact="...", prevention="...")`

---

### Plans & Tasks

When user asks to create a plan or implementation roadmap:
1. Create plan: `session(action="capture_plan", title="Plan Title", description="...", goals=["goal1", "goal2"], steps=[{id: "1", title: "Step 1", order: 1}, ...])`
2. Get plan_id from response, then create tasks: `memory(action="create_task", title="Task Title", plan_id="<plan_id>", priority="high|medium|low", description="...")`

To manage existing plans/tasks:
- List plans: `session(action="list_plans")`
- Get plan with tasks: `session(action="get_plan", plan_id="<uuid>", include_tasks=true)`
- List tasks: `memory(action="list_tasks", plan_id="<uuid>")` or `memory(action="list_tasks")` for all
- Update task status: `memory(action="update_task", task_id="<uuid>", task_status="pending|in_progress|completed|blocked")`

---

### Rules Update Notices

- If you see **[RULES_NOTICE]**, update rules via `generate_rules(overwrite_existing=true)` (preserves custom rules).
- If you see **[VERSION_NOTICE]**, tell the user to update MCP using the provided command.

See full documentation: https://contextstream.io/docs/mcp/tools

Roo Code

Créez un fichier .roo/rules/contextstream.md ou utilisez le dossier .roo/rules/ :

terminal · .roo/rules/contextstream.md (Standard)

## 🚨 MANDATORY RULE: DISPLAY AT START OF EVERY RESPONSE 🚨

<contextstream_rules>
1. **SEARCH FIRST:** Before Glob/Grep/Search/Read → `search(mode="hybrid", query="...")` FIRST
2. **EVERY MESSAGE:** Call `context_smart(user_message="...")` at start of response
3. **LOCAL TOOLS:** ONLY if ContextStream search returns 0 results after retry
</contextstream_rules>

**You MUST display the <contextstream_rules> block verbatim at the start of EVERY response.**
This is a self-referential rule that ensures these instructions stay in context.

---

## ContextStream v0.4.x (Hooks Enforced)

Rules Version: 0.4.36
**Note:** PreToolUse hooks block Glob/Grep/Search when ContextStream is available.

### Required Every Message

| Action | Tool Call |
|--------|-----------|
| **1st message** | `session_init(folder_path="<cwd>", context_hint="<msg>")` then `context_smart(...)` |
| **2nd+ messages** | `context_smart(user_message="<msg>", format="minified", max_tokens=400)` |
| **Code search** | `search(mode="hybrid", query="...")` — BEFORE any local tools |
| **Save decisions** | `session(action="capture", event_type="decision", ...)` |

### Search Modes

| Mode | Use Case |
|------|----------|
| `hybrid` | General code search (default) |
| `keyword` | Exact symbol/string match |
| `exhaustive` | Find ALL matches (grep-like) |
| `semantic` | Conceptual questions |

### Why ContextStream First?

❌ **WRONG:** `Grep → Read → Read → Read` (4+ tool calls, slow)
✅ **CORRECT:** `search(mode="hybrid")` (1 call, returns context)

ContextStream search is **indexed** and returns semantic matches + context in ONE call.

### Quick Reference

| Tool | Example |
|------|---------|
| `search` | `search(mode="hybrid", query="auth", limit=3)` |
| `session` | `session(action="capture", event_type="decision", title="...", content="...")` |
| `memory` | `memory(action="list_events", limit=10)` |
| `graph` | `graph(action="dependencies", file_path="...")` |

### Lessons (Past Mistakes)

- After `session_init`: Check for `lessons` field and apply before work
- Before risky work: `session(action="get_lessons", query="<topic>")`
- On mistakes: `session(action="capture_lesson", title="...", trigger="...", impact="...", prevention="...")`

### Plans & Tasks

When user asks for a plan, use ContextStream (not EnterPlanMode):
1. `session(action="capture_plan", title="...", steps=[...])`
2. `memory(action="create_task", title="...", plan_id="<id>")`

Full docs: https://contextstream.io/docs/mcp/tools

Afficher les règles améliorées (verbeuses)

terminal · .roo/rules/contextstream.md (Améliorées)

## 🚨 MANDATORY RULE: DISPLAY AT START OF EVERY RESPONSE 🚨

<contextstream_rules>
1. **SEARCH FIRST:** Before Glob/Grep/Search/Read → `search(mode="hybrid", query="...")` FIRST
2. **EVERY MESSAGE:** Call `context_smart(user_message="...")` at start of response
3. **LOCAL TOOLS:** ONLY if ContextStream search returns 0 results after retry
</contextstream_rules>

**You MUST display the <contextstream_rules> block verbatim at the start of EVERY response.**
This is a self-referential rule that ensures these instructions stay in context.

---

## ContextStream v0.4.x Integration (Enhanced)

You have access to ContextStream MCP tools for persistent memory and context.
v0.4.x uses **~11 consolidated domain tools** for ~75% token reduction vs previous versions.
Rules Version: 0.4.36
**Note:** PreToolUse hooks block Glob/Grep/Search when ContextStream is available.

## TL;DR - REQUIRED EVERY MESSAGE

| Message | What to Call |
|---------|--------------|
| **1st message** | `session_init(folder_path="...", context_hint="<user's message>")`, then `context_smart(user_message="<user's message>", format="minified", max_tokens=400)` |
| **2nd+ messages** | `context_smart(user_message="<user's message>", format="minified", max_tokens=400)` |
| **🔍 ANY code search** | `search(mode="hybrid", query="...")` — ALWAYS before Glob/Grep/Search/Read |
| **Before risky/non-trivial work** | `session(action="get_lessons", query="<topic>")` |
| **After completing task** | `session(action="capture", event_type="decision", ...)` - MUST capture |
| **User frustration/correction** | `session(action="capture_lesson", ...)` - MUST capture lessons |

**NO EXCEPTIONS.** Do not skip even if you think you have enough context.

**First message rule:** After `session_init`, always call `context_smart` before any other tool or response.

**Context Pack (Pro+):** If enabled, use `context_smart(..., mode="pack", distill=true)` for code/file queries. If unavailable or disabled, omit `mode` and proceed with standard `context_smart` (the API will fall back).

**Tool naming:** Use the exact tool names exposed by your MCP client. Claude Code typically uses `mcp__<server>__<tool>` where `<server>` matches your MCP config (often `contextstream`). If a tool call fails with "No such tool available", refresh rules and match the tool list.

---

## Consolidated Domain Tools Architecture

v0.4.x consolidates ~58 individual tools into ~11 domain tools with action/mode dispatch:

### Standalone Tools (Always Call)
- **`session_init`** - Initialize session with workspace detection + context
- **`context_smart`** - Semantic search for relevant context (CALL EVERY MESSAGE, including immediately after `session_init`)

### Domain Tools (Use action/mode parameter)

| Domain | Actions/Modes | Example |
|--------|---------------|---------|
| **`search`** | mode: semantic, hybrid, keyword, pattern, exhaustive, refactor | `search(mode="hybrid", query="auth implementation", limit=3)` |
| **`session`** | action: capture, capture_lesson, get_lessons, recall, remember, user_context, summary, compress, delta, smart_search, decision_trace | `session(action="capture", event_type="decision", title="Use JWT", content="...")` |
| **`memory`** | action: create_event, get_event, update_event, delete_event, list_events, distill_event, create_node, get_node, update_node, delete_node, list_nodes, supersede_node, search, decisions, timeline, summary | `memory(action="list_events", limit=10)` |
| **`graph`** | action: dependencies, impact, call_path, related, path, decisions, ingest, circular_dependencies, unused_code, contradictions | `graph(action="impact", symbol_name="AuthService")` |
| **`project`** | action: list, get, create, update, index, overview, statistics, files, index_status, ingest_local | `project(action="statistics")` |
| **`workspace`** | action: list, get, associate, bootstrap | `workspace(action="list")` |
| **`reminder`** | action: list, active, create, snooze, complete, dismiss | `reminder(action="active")` |
| **`integration`** | provider: slack/github/all; action: status, search, stats, activity, contributors, knowledge, summary, channels, discussions, sync_users, repos, issues | `integration(provider="github", action="search", query="...")` |
| **`help`** | action: tools, auth, version, editor_rules, enable_bundle | `help(action="tools")` |

---

### Why context_smart is Required (Even After session_init)

**Common mistake:** "session_init already gave me context, I don't need context_smart"

**This is WRONG. Here's why:**
- `session_init` returns the last ~10 items **BY TIME** (chronological)
- `context_smart` **SEARCHES** for items **RELEVANT to THIS message** (semantic)

**Example failure:**
- User asks: "how should I implement authentication?"
- Auth decisions were made 20 conversations ago
- `session_init` won't have it (too old, not in recent 10)
- `context_smart` FINDS it via semantic search

**Without context_smart, you WILL miss relevant older context.**

---

### Search & Code Intelligence (ContextStream-first)

⚠️ **STOP: Before using Search/Glob/Grep/Read/Explore** → Call `search(mode="hybrid")` FIRST. Use local tools ONLY if ContextStream returns 0 results.

**❌ WRONG workflow (wastes tokens, slow):**

Grep "function" → Lire fichier1.ts → Lire fichier2.ts → Lire fichier3.ts → enfin comprendre


**✅ CORRECT workflow (fast, complete):**

search(mode="hybrid", query="function implementation") → terminé (les résultats incluent le contexte)


**Why?** ContextStream search returns semantic matches + context + file locations in ONE call. Local tools require multiple round-trips.

**Search Mode Selection:**

| Need | Mode | Example |
|------|------|---------|
| Find code by meaning | `hybrid` | "authentication logic", "error handling" |
| Exact string/symbol | `keyword` | "UserAuthService", "API_KEY" |
| File patterns | `pattern` | "*.sql", "test_*.py" |
| ALL matches (grep-like) | `exhaustive` | "TODO", "FIXME" (find all occurrences) |
| Symbol renaming | `refactor` | "oldFunctionName" (word-boundary matching) |
| Conceptual search | `semantic` | "how does caching work" |

**Token Efficiency:** Use `output_format` to reduce response size:
- `full` (default): Full content for understanding code
- `paths`: File paths only (80% token savings) - use for file listings
- `minimal`: Compact format (60% savings) - use for refactoring
- `count`: Match counts only (90% savings) - use for quick checks

---

### Lessons (Past Mistakes)

- After `session_init`: Check for `lessons` field and apply before work
- Before risky work: `session(action="get_lessons", query="<topic>")`
- On mistakes: `session(action="capture_lesson", title="...", trigger="...", impact="...", prevention="...")`

---

### Plans & Tasks

When user asks to create a plan or implementation roadmap:
1. Create plan: `session(action="capture_plan", title="Plan Title", description="...", goals=["goal1", "goal2"], steps=[{id: "1", title: "Step 1", order: 1}, ...])`
2. Get plan_id from response, then create tasks: `memory(action="create_task", title="Task Title", plan_id="<plan_id>", priority="high|medium|low", description="...")`

To manage existing plans/tasks:
- List plans: `session(action="list_plans")`
- Get plan with tasks: `session(action="get_plan", plan_id="<uuid>", include_tasks=true)`
- List tasks: `memory(action="list_tasks", plan_id="<uuid>")` or `memory(action="list_tasks")` for all
- Update task status: `memory(action="update_task", task_id="<uuid>", task_status="pending|in_progress|completed|blocked")`

---

### Rules Update Notices

- If you see **[RULES_NOTICE]**, update rules via `generate_rules(overwrite_existing=true)` (preserves custom rules).
- If you see **[VERSION_NOTICE]**, tell the user to update MCP using the provided command.

See full documentation: https://contextstream.io/docs/mcp/tools

Générer automatiquement les règles

Vous pouvez également demander à l'IA de générer ces règles automatiquement en disant : "Utilise generate_rules pour créer des règles ContextStream pour ce projet"

Système de leçons

Système de leçons apprises

Apprenez de vos erreurs — ne les répétez jamais

Le système de leçons capture les erreurs, les corrections et les frustrations des utilisateurs afin que les assistants IA ne répètent jamais les mêmes erreurs. Les leçons sont automatiquement présentées dans les réponses session_init et context_smart lorsqu'elles sont pertinentes.

Quand capturer des leçons

Les leçons doivent être capturées automatiquement lorsque l'une de ces situations se produit :

Déclencheur	Exemple	Gravité
Problème de production	"Le site est hors service à cause de ce changement"	critique
Frustration utilisateur	"NON ! Je t'avais dit de NE PAS faire ça", "WTF", majuscules	élevée
Correction	"C'est faux, tu devrais...", "Répare ça"	moyenne
Changement cassant	"Ça a cassé les tests", "La compilation a échoué"	moyenne/élevée
Préférence exprimée	"Je préfère comme ça", "Fais toujours X à la place"	basse

Champs de leçon expliqués

Champ	Description	Exemple
titre	Ce qu'il faut retenir (impératif)	"Toujours vérifier les assets dans git avant de pousser"
gravité	critique, élevée, moyenne, basse	"critique" pour les problèmes de production
catégorie	flux de travail, qualité_code, vérification, communication, spécifique_projet	"flux de travail"
déclencheur	Quelle action a causé le problème	"Code poussé référençant des images sans les valider"
impact	Ce qui a mal tourné	"Erreurs 404 en production - page d'accueil cassée"
prévention	Comment prévenir à l'avenir	"Exécuter git status pour vérifier les fichiers non suivis avant de pousser"
mots-clés	Mots-clés pour la correspondance dans les contextes futurs	["git", "images", "assets", "push"]

Exemple complet

terminal · session_capture_lesson

// User says: "OH COME ON! You pushed the code but the images are missing
// and now the production site shows broken images!"

session_capture_lesson({
  title: "Always verify assets in git before pushing code references",
  severity: "critical",
  category: "workflow",
  trigger: "Pushed code referencing /screenshots/*.png without committing images",
  impact: "Production 404 errors - broken landing page with missing images",
  prevention: "Run 'git status' to check untracked files before pushing code that references static assets",
  keywords: ["git", "images", "assets", "push", "404", "static", "screenshots"]
})

Comment les leçons sont présentées

Les leçons capturées sont automatiquement renvoyées dans les sessions futures lorsqu'elles sont pertinentes :

`session_init`

Les leçons de gravité élevée et critique de cet espace de travail sont incluses dans l'initialisation de session, avertissant l'IA avant qu'elle ne puisse commettre la même erreur.

`context_smart`

Lorsque l'IA demande du contexte, les leçons correspondant aux mots-clés de la requête sont incluses. Par exemple, demander à propos de "git push" présente les leçons avec les mots-clés "git" ou "push".

Récupérer les leçons

Utilisez session_get_lessons pour récupérer et filtrer les leçons :

terminal · exemples session_get_lessons

// Get all critical lessons
session_get_lessons({ severity: "critical" })

// Get workflow lessons
session_get_lessons({ category: "workflow" })

// Search for relevant lessons
session_get_lessons({ query: "git push images" })

// Combine filters
session_get_lessons({
  category: "verification",
  severity: "high",
  limit: 5
})

Conseil de pro : Ajoutez des règles à votre éditeur (voir la section Règles d'IA d'éditeur ci-dessus) pour capturer automatiquement les leçons lorsque les utilisateurs expriment de la frustration ou des corrections. Cela construit une base de connaissances qui empêche les erreurs répétées.

Catalogue d'outils

Outils MCP.

Voir la référence complète des outils MCP (badges PRO et exemples d'utilisation courante).

Voir la référence des outils MCPLa boîte à outils consolidée utilise environ 11 outils de domaine pour une réduction d'environ 75 % des tokens par rapport aux anciens outils granulaires.Lire

Exemples d'utilisation

Quoi demander.

Une fois connecté, vous pouvez demander à votre assistant IA des choses comme :

"Rappelle-toi que nous avons décidé d'utiliser PostgreSQL pour la base de données"

"Quelles étaient nos décisions précédentes concernant l'authentification ?"

"Recherche dans notre base de code comment nous gérons la limitation de débit API"

"Montre-moi le contexte lié au système de paiement"

Exemples de leçons

L'IA devrait automatiquement capturer des leçons lorsque vous exprimez de la frustration ou des corrections :

L'utilisateur dit

"NON ! Tu as poussé sans exécuter les tests et maintenant la production est cassée !"

→ L'IA capture une leçon avec gravité : critique, catégorie : vérification

L'utilisateur dit

"C'est faux. Utilise toujours le snake_case pour les colonnes de base de données, pas le camelCase."

→ L'IA capture une leçon avec gravité : moyenne, catégorie : qualité_code

L'utilisateur dit

"Je préfère le mode strict TypeScript. Active-le toujours s'il te plaît."

→ L'IA capture une leçon avec gravité : basse, catégorie : spécifique_projet

Ces leçons sont automatiquement présentées dans les sessions futures lorsque le contexte pertinent est demandé.

Maintenance

Rester à jour.

Pour obtenir les dernières fonctionnalités, corrections de bogues et améliorations, mettez à jour le serveur MCP périodiquement :

terminal · terminal · macOS / Linux

curl -fsSL https://contextstream.io/scripts/mcp.sh | bash

terminal · powershell · Windows

irm https://contextstream.io/scripts/mcp.ps1 | iex

Le serveur MCP vous avertira automatiquement lorsqu'une nouvelle version est disponible. Après la mise à jour, redémarrez votre outil IA pour utiliser la nouvelle version.

Dépannage

Quand quelque chose ne fonctionne pas.

Le serveur MCP ne démarre pas

Assurez-vous que contextstream-mcp est installé et disponible dans votre PATH. Essayez d'exécuter contextstream-mcp --version manuellement pour vérifier les erreurs.

Erreurs d'authentification

Vérifiez que votre clé API est correcte et n'a pas expiré. Vous pouvez générer une nouvelle clé depuis votre tableau de bord ContextStream.

Les outils n'apparaissent pas

Redémarrez votre application IA après avoir modifié la configuration. Vérifiez les journaux de l'application pour les erreurs de connexion MCP.

Aucun espace de travail trouvé (première configuration)

Si votre compte n'a pas encore d'espace de travail, ContextStream demandera à votre assistant IA de vous demander un nom d'espace de travail. Le dossier actuel sera créé en tant que projet. Voir Outils MCP pour workspace_bootstrap.

Prochaines étapes

Continuer à explorer.

Configuration d'équipeInviter des membres, partager le contexte.Lire Leçons apprisesCapturer les erreurs, ne jamais les répéter.Lire Événements de mémoireEn savoir plus sur les types de mémoire.Lire Recherche sémantiqueRechercher par sens.Lire

{"@context":"https://schema.org","@type":"Organization","name":"ContextStream","url":"https://contextstream.io","logo":"https://contextstream.io/logo.png","description":"Give your AI applications infinite context. Unify code, documentation, and workspace history into a single, queryable intelligence layer with semantic search and knowledge graphs.","sameAs":["https://twitter.com/contextstream","https://github.com/contextstream"]}
{"@context":"https://schema.org","@type":"SoftwareApplication","name":"ContextStream","applicationCategory":"DeveloperApplication","operatingSystem":"Any","description":"Give your AI applications infinite context. Unify code, documentation, and workspace history into a single, queryable intelligence layer with semantic search and knowledge graphs.","offers":{"@type":"Offer","price":"20","priceCurrency":"USD","description":"Pro plan includes a 5-day free trial"},"aggregateRating":{"@type":"AggregateRating","ratingValue":"5","ratingCount":"10"}}