ashlr-stack
Open-source AI coding stack — bundled MCP servers, agent runtime, and developer tooling for shipping AI-native dev tools.
Ashlr Stack
The control plane for your entire dev stack. One command to provision, wire, and operate every third‑party service in your project.
Status: pre‑alpha, active development.
Install
Four ways, pick one:
# Homebrew (macOS / Linux — installs Phantom Secrets as a dependency)
brew install ashlrai/ashlr/stack
# One-liner, macOS / Linux (also installs Phantom Secrets if missing)
curl -fsSL stack.ashlr.ai/install.sh | bash
# One-liner, Windows (PowerShell)
irm https://stack.ashlr.ai/install.ps1 | iex
# npm / bun registry
bun add -g @ashlr/stack ashlr-stack-mcp # or: npm i -g
Dev install (from a local clone of this repo):
git clone https://github.com/ashlrai/ashlr-stack && cd ashlr-stack
bun install
bun run packages/cli/src/index.ts --help
# Optional: alias stack=`bun run $(pwd)/packages/cli/src/index.ts` so it's on your PATH
Using Stack with an AI coding agent? See STACK.md — a self-contained project brief for agents pulling context from the repo.
What Stack does
In a Claude‑Code‑native world, the friction in starting a project isn't writing code — it's tab‑hopping to create and wire ten services. Every npx create-next-app is followed by an hour of:
- creating a Supabase project, copying URL + anon key + service role key into
.env - generating a Vercel token
- picking a Neon region and copying the connection string
- setting up a Sentry project and pasting the DSN
- registering an OAuth app, generating a PAT, pasting keys, adding MCP servers to
.mcp.json…
Stack collapses that hour into one command:
stack init --template nextjs-supabase-posthog-sentry
# Stack does the OAuth dance per provider,
# creates the upstream resource,
# stores every secret in Phantom,
# writes .env + .mcp.json,
# and hands you a project ready for `bun dev`.
How it relates to the rest of Ashlr
- Phantom Secrets — the vault. Real secret values never leave your machine. Stack writes every credential through Phantom.
- ashlr-plugin — token-efficiency layer for Claude Code. Orthogonal to Stack.
- ashlrcode — multi-provider AI coding CLI. Orthogonal.
Stack is the control plane. Phantom is the vault. ashlr-plugin is the context compressor. They compose.
Curated v1 provider catalog
Database — Supabase · Neon · Turso · Convex · Upstash · Firebase Deploy — Vercel · Railway · Fly.io · Cloudflare · Render Cloud — AWS AI — OpenAI · Anthropic · xAI · DeepSeek Analytics — PostHog Errors — Sentry Payments — Stripe Code — GitHub Tickets — Linear Email — Resend Auth — Clerk
29 providers total. Run stack providers to see the live catalog.
Usage
stack init # interactive template picker
stack add supabase # OAuth → new project → secrets → .mcp.json
stack providers # full catalog (29 services across 11 categories)
stack doctor --fix # verify every service; re-run setup for anything broken
stack exec -- bun dev # run with Phantom's secret proxy active
AI recommendation layer
Describe what you're building — Stack picks the providers.
stack recommend "B2B SaaS with auth, AI, and payments"
# → ranked list of matching providers with rationales
stack recommend "serverless postgres" --save
# → freezes a Recipe to .stack/recipes/<id>.toml
stack apply <recipe-id>
# → runs `stack add` for each provider + pre-wires Phantom rotating envelopes
# + drops webhook stubs for Stripe / Clerk / Supabase / GitHub
# (add --noWire to opt out of the Phantom auto-wiring)
Inside Claude Code, the same flow is one tool call:
stack_recommend { query: "B2B SaaS with auth + payments", save: true }
stack_apply { recipe_id: "<id>" }
Reasoning happens in Claude — Stack owns the catalog + execution. Outside Claude, stack recommend --synth uses a local SLM (LM Studio on :1234, Ollama on :11434) for rationales. No remote LLM SDKs live in Stack.
Bring Stack to an existing project
Already have a repo with services wired up? You don't start from scratch.
# In an existing repo:
stack scan # detects Supabase / Sentry / OpenAI / etc. from package.json, config files, .env.example
stack scan --auto # scans, then interactively runs `stack add` for each detection
stack import # or: inhale an existing .env straight into Phantom + .stack.toml
# Clone someone else's project:
stack clone github.com/org/repo
# → git clone + scans the checkout + prints next steps
# Across every project on this machine:
stack projects list # everywhere you've used Stack
stack doctor --all # run health check across all registered projects
How git-sharing works
Stack splits its config into two files so you can commit the shape of a stack without leaking per-developer resource IDs:
.stack.toml— committed. Names of services, their secret slots, MCP wirings..stack.local.toml— gitignored automatically.project_id,resource_id, timestamps. Unique to each clone.
Another developer cloning the repo runs stack doctor --fix and Stack re-authenticates / re-provisions each service for them, writing a fresh .stack.local.toml.
Monorepo layout
packages/
core/ — @ashlr/stack-core — shared logic, provider adapters
cli/ — @ashlr/stack — the `stack` binary
mcp/ — ashlr-stack-mcp — MCP wrapper
plugin/ — Claude Code plugin wrapper
site/ — Astro landing page (deploys to stack.ashlr.ai)
templates/ — starter stacks
docs/ — auth matrix, schema reference
Publishing
Three packages ship to npm: @ashlr/stack-core, ashlr-stack-mcp, @ashlr/stack. For monorepo dev, @ashlr/stack depends on @ashlr/stack-core via workspace:* — that's what lets bun install link the local checkout. A plain npm install @ashlr/stack from outside the workspace cannot resolve workspace:*, so the publish flow has to rewrite those ranges to an actual version (e.g. ^0.1.0) right before npm publish.
Don't hand-edit the workspace:* entries in packages/*/package.json — dev needs them. Use the publish script instead:
scripts/publish.sh --version 0.1.0
It bumps each package's version, swaps workspace:* → ^<version>, runs npm publish --dry-run for verification, asks for explicit confirmation, publishes in dep order (core → mcp → cli), restores workspace:* so local dev keeps working, and tags the release. See the script header for details.
Landing page
cd packages/site
bun install
bun run dev # http://localhost:4321
bun run build # static output in dist/
Dark-first, magenta accent, Astro + Tailwind v4 + Framer Motion. Three interactive React islands (animated terminal, "with vs without Stack" tab comparison, Claude Code chat mock). prefers-reduced-motion honored.
License
MIT.
Verwandte Server
Alpha Vantage MCP Server
SponsorAccess financial market data: realtime & historical stock, ETF, options, forex, crypto, commodities, fundamentals, technical indicators, & more
ZenML
Interact with your MLOps and LLMOps pipelines through your ZenML MCP server
iOS MCP Server
An iOS mobile automation server using Appium and WebDriverAgent.
env-secret-exposure-analyzer-mcp
Scans projects for secret exposure risks in .env files and logs
ClipToWSL
Enables AI coding agents to read Windows clipboard contents, including text and images, from within the Windows Subsystem for Linux (WSL).
MCP JS Debugger
Debug JavaScript and TypeScript applications through the Chrome DevTools Protocol with full source map support.
Dev Tools for Android (DTA) MCP
MCP server which expose Android's app runtimes to coding agents.
Docker MCP server
Manage Docker containers, volumes, and services using natural language commands.
agency-mcp-server
On-demand access to 150+ specialist AI agent templates — search, browse, and spawn agents. 150x reduction in context usage vs loading agents locally.
Airflow MCP Server
Control Apache Airflow via its API using JWT authentication.
Just Prompt
A unified interface for various Large Language Model (LLM) providers, including OpenAI, Anthropic, Google Gemini, Groq, DeepSeek, and Ollama.