# MCP Server Scan by Agentropolis Scan a live remote MCP server before your agent connects to it. Give the server URL (or paste its tools/list): the scan reads initialize + tools/list only (never calls a tool) and flags poisoned or hidden tool instructions: / tags, "do not tell the user", asks to read secrets or stuff the conversation into a parameter, cross-tool shadowing, invisible Unicode, HTML-comment instructions and payment injection, plus suspicious parameter names and duplicate tool names. Every tool is fingerprinted (sha256 of name, description, inputSchema, annotations); pass the fingerprint from your last scan to catch silent tool changes (MCP "rug pulls"). Rules-only (same rules as Skill Scan and the Payment-Injection Screen). Returns allow / review / block with evidence. Advisory. Price: $0.03 USDC per call on Base mainnet. No free calls; free sample output: https://seenly-skill-scan.vercel.app/samples/mcp-scan-sample.json POST https://seenly-skill-scan.vercel.app/api/mcp-scan Body: {"url":"https:///mcp"} or {"tools":}; optional "baseline": the "fingerprint" object from an earlier scan. Returns: verdict allow|review|block, risk 0-100, reasons, findings with tool + evidence, suspicious_params, fingerprint {server, tools{name: sha256}}, changes {added, removed, changed} when a baseline is passed. Unpaid calls return HTTP 402 with x402 v2 PAYMENT-REQUIRED; settlement only on 2xx. OpenAPI: https://seenly-skill-scan.vercel.app/mcp/openapi.json · Kit: https://seenly-skill-scan.vercel.app/mcp/kit.json · Contact: hello@agentropolis.io · https://agentropolis.io