{"name":"TraceCheck","description":"Independent evaluation service for AI agent execution traces. Submit a goal, trace, and outcome to receive a reliability score (0-100), per-dimension sub-scores, detected issues and concrete process-improvement suggestions. No content needed: send step types only (privacy-light format); no goal/trace/outcome text is logged or stored. Free during test phase, no API key. Designed for agents that want to verify their own work or improve long-horizon reliability. Also available as an MCP server (streamable HTTP) at https://tracecheck-mcp.onrender.com/mcp with tools evaluate_trace and get_trace_schema.","version":"0.4.0","provider":{"organization":"TraceCheck (test)","url":"https://tracecheck-mcp.onrender.com"},"documentationUrl":"https://github.com/TGS-checker/tracecheck-mcp#readme","iconUrl":"https://tracecheck-mcp.onrender.com/icon.png","supportedInterfaces":[{"url":"https://tracecheck-mcp.onrender.com/a2a","protocolBinding":"HTTP+JSON","protocolVersion":"1.0"},{"url":"https://tracecheck-mcp.onrender.com/a2a","protocolBinding":"JSONRPC","protocolVersion":"0.3"}],"capabilities":{"streaming":false,"pushNotifications":false,"extendedAgentCard":false},"defaultInputModes":["application/json","text/plain"],"defaultOutputModes":["application/json"],"skills":[{"id":"evaluate_trace","name":"Evaluate agent execution trace","description":"Score an agent's multi-step execution trace for reliability and return concrete improvement suggestions. No content needed: send step types only. Send POST /a2a/message:send (or JSON-RPC message/send at POST /a2a) with one message part {\"data\": {\"goal\": str, \"trace\": str | list, \"outcome\": str, \"agent_name\": str (optional)}}. Privacy-light (recommended): goal = a generic category ('data lookup'), outcome = 'success'|'partial'|'failed', trace = step types + statuses only, e.g. [{\"type\":\"action\",\"tool\":\"search\",\"status\":\"ok\"},{\"type\":\"error\"},{\"type\":\"retry\"},{\"type\":\"verify\",\"status\":\"passed\"}] or 'action:search ok > error > retry > verify passed'. Free text also works: chronological steps, one per line, with what each tool returned, errors with their recovery, and a final 'Verified:' line. Nothing you send is logged or stored, only the score and counts. Returns a Message whose data part is {score, dimensions, issues, suggestions, summary, privacy_mode, version}. Loop: run task, evaluate, apply suggestions, re-run, compare.","tags":["evaluation","agent-improvement","verification","self-improvement","trace","reliability","privacy"],"examples":["Evaluate this trace for the goal 'book a flight and confirm the reservation'.","Check if my multi-step research process completed correctly and suggest improvements.","{\"goal\": \"data lookup\", \"trace\": \"action:search ok > action:fetch_details error > retry:fetch_details ok > verify passed\", \"outcome\": \"success\"}","{\"goal\": \"Find the cheapest direct flight AMS->LIS on 12 March and report price and flight number\", \"trace\": \"1. Called search_flights(...)\\n2. Observed: returned 4 flights...\\n3. Verified: matches goal\", \"outcome\": \"KL1693, EUR 89\"}"],"inputModes":["application/json","text/plain"],"outputModes":["application/json"]}]}