MCP Server Scan by Agentropolis

Check a live MCP server before your agent connects to it. Give the server URL (or paste its tools/list). The scan reads initialize and tools/list only, never calls a tool, and flags poisoned or hidden tool instructions: <IMPORTANT> tags, "don't tell the user", asks to read secrets or stuff the conversation into a parameter, cross-tool shadowing, invisible Unicode, comment-hidden text and payment injection.

Fingerprints: every tool gets a sha256 of its name, description, input schema and annotations. Save the fingerprint and pass it back as baseline next time: added, removed or changed tools are reported, so a server that quietly swaps its tool text after you approved it (an MCP "rug pull") gets caught.

$0.03 USDC per scan (LIVE: real USDC payments). Rules-only (the same rules as Skill Scan and the Payment-Injection Screen), no LLM. Not charged if the server can't be reached. Advisory. For a full source-code scan use Skill Scan.

curl -s https://seenly-skill-scan.vercel.app/api/mcp-scan -H 'content-type: application/json' \
  -d '{"url":"https://mcp.example.com/mcp"}'
# -> HTTP 402 + PAYMENT-REQUIRED (x402 v2). Pay, retry with PAYMENT-SIGNATURE.
# Re-scan later: {"url":"...","baseline":<fingerprint from last time>}
# Auth-only server: {"tools":{"tools":[...]}}  (paste its tools/list)

Machine-readable: openapi.json · llms.txt · kit.json · x402 · sample output

Agentropolis badge First-party Agentropolis kit.

Agentropolis kit · listed on Agentropolis · contact hello@agentropolis.io