XLUXX Trust Layer

Runtime trust scoring for the Model Context Protocol. Makes AI agents reliable.

XLUXX is a runtime trust layer for AI agents that use MCP (Model Context Protocol) tools. Before an agent calls any tool, XLUXX tells it which one is most reliable — with confidence scores, fallbacks, risk flags, and the Resonance Engine for temporal stability analysis. The Context Gate prevents agents from returning responses when their context has drifted from its original instructions.

• 15,000+ MCP servers indexed
• 7,600+ tools cataloged
• 100% scoring coverage — every server has a fresh trust score
• Continuous testing every 15 minutes

---

Table of contents

• Quick start
• Why XLUXX
• The Resonance Engine
• The Context Gate
• API reference
• SDKs
• Pricing
• Links

---

Quick start

Curl

curl -X POST https://api.xluxx.net/resolve-tool \
  -H "Content-Type: application/json" \
  -d '{"intent": "search the web"}'

{
  "best_server": "brave-search",
  "confidence": 0.985,
  "resonance_score": 98.5,
  "fractal_reliability": 100,
  "reason": "excellent uptime, fast response, consistent",
  "fallback": "fetch",
  "risk_flags": []
}

Python

pip install xluxx

from xluxx import TrustLayer

trust = TrustLayer()
result = trust.resolve("search the web")
print(result.best_server, result.confidence)

Node.js

npm install xluxx-trust

const { resolveTool } = require('xluxx-trust');
const result = await resolveTool('search the web');
console.log(result.best_server, result.confidence);

MCP server (Claude Desktop, Cursor, etc.)

npx -y xluxx-mcp-server

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "xluxx": {
      "command": "npx",
      "args": ["-y", "xluxx-mcp-server"]
    }
  }
}

---

Why XLUXX

AI agents fail when the tools they call are unreliable. Most MCP servers don't publish uptime, latency, or schema-stability data — and even when they do, the agent has no way to act on it at runtime.

XLUXX sits in the agent's execution path. It:

1. Resolves intents to tools — given "search the web", it returns the most reliable MCP server for that capability.
2. Scores every tool continuously — uptime, latency, schema drift, response coherence.
3. Surfaces risks before they bite — bursty failures, degradation cascades, behavioral drift.
4. Provides fallbacks — independent backups, ranked by capability overlap.
5. Gates the agent's output — blocks responses when the agent has lost track of its original instructions.

---

The Resonance Engine

XLUXX doesn't just check whether a tool is up. It measures whether the tool remains stable, coherent, and safe to compose into agent workflows. Four metrics:

Metric	What it measures
Fractal Reliability Index	Detects bursty failures, degradation cascades, and unstable latency patterns across multiple time scales
Coherence Drift	Whether a server's behavior is changing from its baseline (schema drift, response inconsistency, capability shift)
Toolchain Resonance	Scores tool combinations, not just individuals — exposes weakest links and risk-adjusted reliability for multi-step chains
Fallback Stability	Ensures fallbacks are independent and reliable — a fallback that fails when the primary fails is no fallback

---

The Context Gate

A context firewall for AI agents. Long conversations cause agents to drift from their original instructions — through context window pressure, prompt injection, or simple cumulative noise. The Context Gate detects this and blocks responses when drift exceeds a threshold.

Workflow:

1. Register the original system prompt at the start of a session → get a session_id
2. Check between turns to monitor drift continuously (gate=PASS or gate=BLOCK)
3. Verify before returning critical responses to the user as a final gate
4. Inspect the audit trail of all checks for the session

When the gate blocks, the response includes the missing instructions and recommended re-anchoring text.

---

API reference

Base URL: https://api.xluxx.net

All endpoints accept an optional X-API-Key header. Free tier works without a key (rate-limited). Higher tiers require a key from /register.

Core

Endpoint	Method	Description
/health	GET	Service health and version
/	GET	API root with full endpoint listing
/resolve-tool	POST	Find the best tool for an intent
/rank-tools	POST	Rank a list of tools by trust score
/recommend	POST	Generate tool-chain recommendations for a task
/servers	GET	List all tracked servers (?limit=)
/servers/:id	GET	Get a specific server's full record
/search	GET	Search servers by capability (?q=)
/categories	GET	List server categories with counts
/trending	GET	Trending servers by recent activity
/evaluate	POST	Submit a server URL for live evaluation
/status/:id	GET	Real-time server health and outage detection
/trends/:id	GET	Historical reliability (24h, 7d, all-time)
/stats	GET	Aggregate platform stats
/tools	GET	List all cataloged tools
/alerts	GET	Active alerts for degraded or down servers

Resonance Engine

Endpoint	Method	Description
/resonance/:id	GET	Fractal Reliability + Coherence Drift for a server
/resonance/:id/predict	GET	Predict near-term reliability (?horizon= hours)
/resonance/:id/drift	GET	Detect coherence drift (?window=24h|7d|30d)
/resonance/:id/report	GET	Comprehensive trust report (markdown or JSON)
/resonance/:id/explain	GET	Factor breakdown explaining a server's score
/resonance/compare	POST	Compare a primary against alternatives
/resonance/fallback	POST	Find best independent fallback for a server
/resonance/fleet	POST	Aggregate health across a fleet of servers
/resonance/chain	POST	Score a multi-step toolchain

Context Gate

Endpoint	Method	Description
/context-gate	POST	One-shot drift check (no session)
/context-gate/register	POST	Register a session, get session_id
/context-gate/check	POST	Continuous mid-conversation check
/context-gate/verify	POST	Final gate before returning a response
/context-gate/session/:id	GET	Inspect a session and its check history
/context-gate/session/:id	DELETE	Close a session

SEO and discovery

Endpoint	Method	Description
/seo/server-pages	GET	Generate SEO-friendly per-server pages
/seo/category-pages	GET	Generate SEO-friendly per-category pages
/badge/:id.svg	GET	Trust badge SVG for a server
/hub/:id	GET	Public hub page for a server
/mcp/:id	GET	MCP-server metadata page

Account and management

Endpoint	Method	Description
/register	POST	Get a free API key
/reissue-key	POST	Reissue an API key
/upgrade/:tier	GET	Upgrade flow for Starter / Pro
/usage	GET	Current usage and quota
/analytics	GET	API usage analytics

MCP and ingestion

Endpoint	Method	Description
/.well-known/mcp.json	GET	MCP discovery manifest
/import	POST	Bulk-import server records
/import-tools	POST	Bulk-import tool records
/crawl-tools	POST	Trigger tool crawler
/score-all	POST	Re-score every tracked server

---

SDKs

SDK	Package	Install
Python	xluxx	pip install xluxx
Node.js / TypeScript	xluxx-trust	npm install xluxx-trust
MCP Server	xluxx-mcp-server	npx -y xluxx-mcp-server
LangChain resolver	bundled in repo	see sdk/langchain/xluxx_resolver.py

The MCP server exposes 16 tools: resolve_tool, rank_tools, recommend_toolchain, check_server_status, get_reliability_trends, get_resonance_score, search_servers, get_alerts, analyze_toolchain_risk, predict_reliability, compare_alternatives, detect_drift, generate_trust_report, find_fallback, monitor_fleet, explain_score, verify_context_integrity, verify_context. See mcp-server/README.md for full documentation.

---

Pricing

Tier	Price	Calls / day	Use case
Free	$0	100	Personal projects, evaluation
Starter	$49 / mo	1,000	Small production agents
Pro	$149 / mo	10,000	High-volume agent fleets
Enterprise	Custom	100,000+	SLA, on-prem, custom scoring

Get a free key:

curl -X POST https://api.xluxx.net/register \
  -H "Content-Type: application/json" \
  -d '{"name": "your-name", "email": "[email protected]"}'

---

Links

• Website: https://xluxx.net
• API: https://api.xluxx.net
• npm (SDK): https://www.npmjs.com/package/xluxx-trust
• npm (MCP server): https://www.npmjs.com/package/xluxx-mcp-server
• PyPI: https://pypi.org/project/xluxx/
• GitHub: https://github.com/DrDMT-VR/xluxx-trust

---

License

Apache-2.0

Author

Pablo Solorzano Cohen — [email protected]

---

Keywords: MCP, Model Context Protocol, AI agent reliability, MCP trust scoring, MCP server registry, agent runtime, tool selection, context gate, context firewall, prompt injection defense, resonance engine, fractal reliability, coherence drift, toolchain resonance, LLM observability.