mcp-scan — vet an MCP server before you connect an agent to it
Static analysis of a server's advertised tool metadata (descriptions, schemas, hidden unicode). It does not call the server's tools. Free preview returns tools scanned, issue counts by severity, risk score and verdict (clear / review / block).
Remote server (Streamable HTTP)
B=https://x402-data-api.sigrunner.workers.dev
curl -s "$B/scan/mcp/preview?url=https://HOST/mcp"
Private, OAuth-gated or stdio server
Paste its tools/list result; nothing needs to be reachable from outside:
curl -s -X POST "$B/scan/mcp/preview" -H 'content-type: application/json' \
-d '{"tools":[ ...the "tools" array from tools/list... ]}'
Acting on the verdict
clear— proceed.review— read the flagged tool descriptions before enabling the server.block— do not connect; tell the user which tool and why.
The full itemized report with evidence and OWASP mapping is the paid route GET /scan/mcp ($0.05 USDC via x402 on Base). CLI: npx github:rezearcher/greyridge-x402-tools mcpscan <url> (exit code 2 = block). Methodology and measured limits: https://github.com/rezearcher/x402-data-api/blob/main/docs/STATUS.md