Skip to main content
MCP server CX Codex

litmus — Claude Code MCP server: grade servers A to F

Grades MCP servers and skills using the open behavioral litmus methodology, A to F. Local install runs the full probe suite; hosted endpoint serves lookups only.

Install into Claude Code · Platform CLI: remote server All methods and platforms ↓
claude mcp add --transport http litmus https://polygraph.so/api/mcp
Open on GitHub

What litmus does

The server ships six tools. check_server reads a published grade in under a second, no code execution. run_litmus runs the full harness: connects as an agent would, fingerprints the tool surface, runs four probe categories (C-02 needs Docker; without it the grade caps at B), and returns the grade with an evidence bundle on disk. run_skill_litmus grades Claude Code skills with a static scan. list_servers returns servers with published grades, sorted descending. request_grade submits a paid grading request to polygraph.so. verify_attestation reads the onchain proof behind a published grade in EAS on Base.

The hosted endpoint at https://polygraph.so/api/mcp requires no install and serves three lookup tools: check_server, list_servers, request_grade. The stdio variant via npx -y @polygraphso/litmus installs the full toolset, including run_litmus and run_skill_litmus, which execute the target server's code. For Claude Code there is a plugin with two slash commands: /polygraph:grade <server> and /polygraph:check <server>. The repository also includes a GitHub Action polygraphso/litmus@v1 that fails the build when an MCP server or skill grades D or F.

What it helps with

  • check an MCP server before install using the open methodology
  • grade a server or skill A to F directly from your agent
  • list servers with published grades, sorted by quality
  • fail a CI build when a dependency grades D or F
  • read the blockchain attestation of a published grade on Base

Example prompts for the agent

  • Check the grade for npm/@modelcontextprotocol/server-filesystem in the polygraph.so index
  • Run the full litmus harness for github/owner/repo and return the grade with evidence
  • Show me the first 10 MCP servers with grade A

How to install litmus

1Platform CLI: remote server

claude mcp add --transport http litmus https://polygraph.so/api/mcp

2Platform CLI

claude mcp add --env POLYGRAPH_API_URL=<POLYGRAPH_API_URL> --env NEXT_PUBLIC_POLYGRAPH_NETWORK=<NEXT_PUBLIC_POLYGRAPH_NETWORK> --transport stdio litmus -- npx -y @polygraphso/litmus

Replace <…> with your own values

Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more

Questions about litmus

How to install litmus in Claude Code?

For the hosted endpoint, run claude mcp add --transport http litmus https://polygraph.so/api/mcp. For the full toolset: `claude mcp add --env POLYGRAPH_API_URL=<POLYGRAPH_API_URL> --env NEXT_PUBLIC_POLYGRAPH_NETWORK=<NEXT_PUBLIC_POLYGRAPH_NETWORK> --transport stdio litmus -- npx -y @polygraphso/litmus. The plugin installs with /plugin marketplace add polygraphso/litmus and /plugin install polygraph@polygraphso`.

What does litmus do?

litmus grades MCP servers and skills using the open behavioral litmus methodology, A to F. The hosted variant looks up published grades in under a second; the stdio variant runs the full harness, executing the target server's code and running four probe categories.

Does litmus require Docker?

Docker is required for the C-02 probe. Without Docker, that probe is skipped and the grade caps at B. The other three probe categories work without Docker.

Similar MCP servers

All MCP servers for Agent setup: memory, context, orchestration →

Project sponsors

Companies that keep this analytics open
Gold sponsor

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor