{
  "kit": "0.1",
  "id": "seenly.mcp-server-scan",
  "version": "0.1.0",
  "name": "MCP Server Scan — check a live MCP server for poisoned tool instructions and silent tool changes",
  "summary": "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: <IMPORTANT>/<SYSTEM> 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. $0.03 USDC.",
  "category": [
    "security_verification",
    "developer_tools"
  ],
  "seller": {
    "name": "Agentropolis",
    "domain": "seenly-skill-scan.vercel.app",
    "url": "https://seenly-skill-scan.vercel.app/mcp",
    "verify": "well-known",
    "contact": "hello@agentropolis.io",
    "test_wallets": [
      "0xcaA54748dA198081A89b6f369d798Ad78FC35B83"
    ]
  },
  "invoke": [
    {
      "kind": "http",
      "method": "POST",
      "url": "https://seenly-skill-scan.vercel.app/api/mcp-scan",
      "openapi": "https://seenly-skill-scan.vercel.app/mcp/openapi.json#/paths/~1api~1mcp-scan/post",
      "content_types": [
        "application/json"
      ]
    }
  ],
  "input_schema": {
    "type": "object",
    "properties": {
      "url": {
        "type": "string",
        "description": "Public https Streamable HTTP MCP endpoint. The scan runs initialize + tools/list only; tools are never called."
      },
      "tools": {
        "description": "Alternative to url: a pasted tools/list result ({\"tools\":[{name, description, inputSchema}]}), e.g. for servers that need auth."
      },
      "baseline": {
        "type": "object",
        "description": "Optional: the \"fingerprint\" object from an earlier scan, to detect added/removed/changed tools."
      },
      "request_key": {
        "type": "string"
      }
    }
  },
  "output_schema": {
    "$ref": "https://seenly-skill-scan.vercel.app/mcp/openapi.json#/components/schemas/McpScanResult"
  },
  "sample": {
    "input": {
      "url": "https://mcp.example.com/mcp",
      "baseline": {
        "server": "...sha256 from your last scan...",
        "tools": {
          "get_weather": "...sha256..."
        }
      }
    },
    "output_url": "https://seenly-skill-scan.vercel.app/samples/mcp-scan-sample.json"
  },
  "offers": [
    {
      "rail": "x402",
      "scheme": "exact",
      "network": "eip155:8453",
      "asset": "USDC",
      "asset_address": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913",
      "amount": "30000",
      "price_usd": "0.03",
      "payTo": "0xeE6239F3E17bEBe5e89F21F29D74D68CAa0d78CE",
      "charge_on": "success",
      "environment": "live",
      "path": "/api/mcp-scan",
      "name": "MCP Server Scan"
    }
  ],
  "trial": {
    "type": "sample",
    "sample_url": "https://seenly-skill-scan.vercel.app/samples/mcp-scan-sample.json",
    "free_calls_per_wallet": 0
  },
  "behavior": {
    "side_effects": "none (initialize + tools/list only; tools are never called)",
    "idempotent": true,
    "deterministic": true,
    "p95_latency_s": 10,
    "advisory_only": true,
    "data_retention": "none",
    "max_payload_kb": 240,
    "rate_limit": "20/min per client IP"
  },
  "limits": {
    "servers": "public https Streamable HTTP MCP endpoints on port 443 (no private IPs, no redirects, no auth); servers that need auth: paste tools/list",
    "tools": "up to 200 tools, 5 tools/list pages, 2 MB reply",
    "bulk": "one server per call"
  },
  "policy": {
    "refund": "Not charged on failure: invalid input is rejected before payment, and unreachable / non-MCP / auth-only servers return an error before settlement.",
    "data": "Tool definitions are processed in memory and not stored. No LLM is used.",
    "advisory": "Advisory static scan of what the server tells agents. A clean result is not a guarantee; servers can change tools later, so keep the fingerprint and re-scan."
  },
  "contact": {
    "email": "hello@agentropolis.io"
  },
  "discovery": {
    "landing": "https://seenly-skill-scan.vercel.app/mcp",
    "openapi": "https://seenly-skill-scan.vercel.app/mcp/openapi.json",
    "llms_txt": "https://seenly-skill-scan.vercel.app/mcp/llms.txt",
    "x402": "https://seenly-skill-scan.vercel.app/.well-known/x402"
  },
  "check_version": "mcp-server-scan-0.1.0"
}