# MCP Judge

MCP Judge is an MCP server that automatically evaluates OTHER MCP servers and returns an objective quality score and readable report, so an AI agent can decide which MCP server to trust and integrate — before wiring it in. An unreliable or undocumented MCP wastes tokens and can break workflows; MCP Judge tests it in seconds.

Method: objective automated checks only — no human reviewer, no LLM opinion. It runs the MCP 'initialize' handshake and 'tools/list', measures latency, scores tool and parameter documentation coverage, and safely probes free tools. It NEVER pays to test a target's paid tools; it only verifies they are cleanly declared. It does not judge the real-world usefulness or safety of content.

MCP endpoint (Streamable HTTP): https://mcp-judge.onrender.com/mcp

Tools:
- quick_check (FREE): instant trust check — 0-100 score, A-F grade, tool count, latency, one-line verdict.
- evaluate_mcp ($0.05): full report — documentation coverage, functional probe, per-criterion score breakdown, plain-language summary.
- compare_mcps ($0.10): 2-5 servers head-to-head — ranking, recommended winner, per-server reports.
- list_criteria (FREE): the exact scoring criteria, weights, and A-F grade scale.

Payment: x402 protocol (USDC micropayments over HTTP 402), network eip155:8453. No account or API key needed — call a paid tool without payment to receive the payment requirements, then retry with the X-PAYMENT header. Client libraries: @x402/fetch, @x402/mcp.

Recommended agent workflow: quick_check (free) -> evaluate_mcp (full report) -> compare_mcps (to pick between alternatives). Scoring is fully transparent via list_criteria.