codetrawl-mcp
v0.2.0
Published
CodeTrawl's deterministic code analysis as MCP tools for AI agents — blast radius, untested hotspots, structural duplicates, health signals, and change simulation on any GitHub repo. Computed, never guessed.
Downloads
342
Maintainers
Readme
codetrawl-mcp
Deterministic code analysis as Model Context Protocol tools, so AI coding agents can ask computed questions about a GitHub repo — blast radius, untested hotspots, structural duplicates, a 17-signal health verdict, and "what does this change break?" simulation — instead of grepping and guessing.
Powered by CodeTrawl. Transport is stdio; the server runs as a child process of your MCP client and analyzes public repos on demand (no data leaves your machine except the repo fetch from GitHub).
Install
claude mcp add codetrawl npx codetrawl-mcpOr globally, then point any MCP client at the codetrawl-mcp binary:
npm install -g codetrawl-mcpGitHub token (recommended): analysis fetches the repo from GitHub. Set
GITHUB_TOKEN in the server's environment to lift the 60-request/hour
unauthenticated limit to 5000/hour.
Tools
analyze_repo— entry point; returns a sessionId every other tool needsblast_radius— what a file or function change reachesuntested_hotspots— complex production functions with no test callerfind_duplicates— structurally identical functions across the reposignals— the full 17-signal health verdict + dimension rollupcompare_sessions— structural diff between two analysesanalyze_diff/review_changes— change-aware reviewsimulate_change— the Conscience: the deterministic blast + gate on a proposed diff, before you commit it
Plus the conscience prompt — the agent loop that runs simulate_change and
respects its gate.
License
PolyForm Noncommercial 1.0.0 — free for noncommercial use. For commercial use, contact coffeejones.
