Mcp graveyard
Audit which MCP server tools your Claude Code sessions actually invoke. Use when the user wants to find dead MCP servers (configured but never called), hallucinated MCP tool calls, or per-project MCP usage. Triggers on "which MCP servers don't I use?", "clean up MCP config", "remove unused MCP servers", "audit MCP tools". Runs locally; reads ~/.claude session JSONL logs and ~/.claude.json. No network calls.From its SKILL.md
npx -y skills add sfrangulov/skill-graveyard --skill mcp-graveyardAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 7 stars7 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
- runs commandsInstructs the agent to run 4 commands, including `npx mcp-graveyard@latest <subcommand> [flags]` and 3 more.
What its file declares
Copied from the file, not written here
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
1.4 KB, 252 tokens by cl100k_base, as published. Nobody here has run it
mcp-graveyard
CLI that audits MCP server tool usage. Server-first table grouped into ACTIVE / DEAD / HALLUCINATED / MISSING buckets.
How to invoke
Run via npx mcp-graveyard@latest <subcommand> [flags] — no install needed.
Subcommands
audit(default) — classify every server. Flags:--days N,--only <bucket>,--tools <server>(drill-in),--json,--claude-dir <path>.prune— print removal plan.--applyto execute (with auto-backup of removed config to a timestamped file).suggest— actionable triage for unused/hallucinated.projects— per-project breakdown.
Decision flow
- "which MCP servers do I not use?" →
npx mcp-graveyard@latest - "clean up MCP config" →
prune, then--apply - "MCP tool didn't work" →
suggest - per-project breakdown →
projects
After running, print output verbatim. Do not summarize unless the user asks.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.