Skip to main content
Lighthouse for MCP — audit any Model Context Protocol server and get a scored, actionable report in seconds. 
uvx mcpscore https://your-server.example/mcp

npx @mcp-box/mcpscore https://your-server.example/mcp

uv tool install mcpscore
mcpscore https://your-server.example/mcp

pip install mcpscore
mcpscore https://your-server.example/mcp
MCP servers that violate the spec fail silently in the worst place: inside someone else’s AI agent. A missing tool description, an outdated protocol version, or an unencrypted endpoint won’t crash your server — it will just make agents pick the wrong tool, drop your server from their registry, or leak traffic. mcpscore catches these issues before your users do.

​
What you get

  • A severity-weighted score across protocol compliance, server metadata, capabilities, tools, security, and transport — deterministic, no API keys, CI-ready.
  • A readiness score for the next spec revision (2026-07-28, stateless lifecycle) — separate from your main score, so it informs without punishing.
  • Actionable messages — every failed check says what to fix, and every rule is anchored to the spec (see the methodology).

​
Quick start

# Local servers (stdio)
mcpscore path/to/server.py
mcpscore path/to/server.js

# Remote servers (Streamable HTTP / SSE, auto-detected —
# including modern-only servers on the 2026 stateless lifecycle)
mcpscore https://example.com/mcp

# Machine-readable report for CI
mcpscore https://example.com/mcp --json > report.json
Example output tail: 
Audit finished. Final score: 88/97
Spec: 2025-11-25 negotiated (latest: 2025-11-25) · era: legacy
Readiness for MCP 2026-07-28: 3/13 (informative — not part of the main score)

​
Learn more

Scoring Methodology

How the score works, and why you can trust it

Rules Reference

Every rule, its severity, and when it applies

GitHub Action

Gate pull requests on quality, with a report comment

Source

mcp-box/mcpscore — issues and contributing

PyPI

uv tool install mcpscore · pip install mcpscore