{"attribution":"MCP Queen — the evidence layer for MCP (https://mcpqueen.com)","license":"CC BY 4.0 — free to use with attribution and a link to mcpqueen.com","methodology":"https://mcpqueen.com/registry#methodology","docs":"https://mcpqueen.com/api","generated_at":"2026-10-27T12:27:53.235Z","server":{"name":"io.github.JcJamet/ia-qa-eval","title":"IA-QA Eval & RAG — 38 tools","description":"LLM & RAG evals: hallucination, eval contracts, semantic tests, tokens & model info. Free, no key.","version":"1.0.1","remote_url":"https://www.ia-qa.com/mcp/eval","repo_url":"https://github.com/ia-qa-com/mcp"},"operational_grade":{"grade":"A","score":90,"provisional":0,"reachable":1,"auth_state":"open","latency_ms":650,"tool_count":38,"probed_at":"2026-10-27T08:15:39.297Z","evidence":[{"criterion":"reachability","points":25,"max":25,"evidence":"HTTP 200, initialize accepted in 650ms"},{"criterion":"protocol","points":15,"max":15,"evidence":"valid JSON-RPC initialize result, protocolVersion 2025-06-18, serverInfo ia-qa-toolbox-eval@1.0.0"},{"criterion":"tooling","points":35,"max":35,"evidence":"tools/list OK: 38 tools e.g. \"get_testing_guidelines\"; 38/38 described, 38/38 fully-typed schemas, median description 275 chars"},{"criterion":"latency","points":7,"max":10,"evidence":"initialize round-trip 650ms"},{"criterion":"provenance","points":8,"max":15,"evidence":"description present; repository linked; version 1.0.1; namespace io.github.JcJamet does NOT match remote host (expected *.jcjamet.github.io)"}],"protocol_access":"open"},"field_semantics":{"auth_state":"Deprecated name retained for compatibility. It means protocol handshake/discovery access only, not tool-level data entitlement.","protocol_access":"Preferred name for auth_state."},"audit_state":"observations_published","observations":[{"dimension":"citation_quality","metric":"citation_capability_advertised","status":"unverified","value_text":null,"evidence":"Metadata/tool descriptions advertise citation or evidence capability: metamorphic_check Reference-free stability primitive: instead of comparing an answer to a ground truth, it checks that an assistant's answer stays INVARIANT when the QUESTION is transform | Catches the failure class no reference answer can expose — an assistant that handles one phrasing well and a trivial variant of it badly | If a `reference` (ground truth) is given, scores each output against it and picks the closer one. No response-level citation benchmark has verified this claim yet.","source_type":"live_probe","sample_size":null,"observed_at":"2026-10-27T08:15:39.297Z","methodology_version":"trust-v1"},{"dimension":"security","metric":"sensitive_capabilities_declared","status":"concern","value_text":"generate_ci_workflow","evidence":"1 discovered tool(s) advertise potentially state-changing or privileged capabilities. Tool names: generate_ci_workflow. Classification is description-based and does not prove exploitability.","source_type":"live_probe","sample_size":1,"observed_at":"2026-10-27T08:15:39.297Z","methodology_version":"trust-v1"},{"dimension":"security","metric":"authentication_boundary","status":"observed","value_text":"open","evidence":"Live protocol probe observed auth_state=open. This describes discovery/access behavior; it is not by itself a vulnerability finding.","source_type":"live_probe","sample_size":null,"observed_at":"2026-10-27T08:15:39.297Z","methodology_version":"trust-v1"}],"response_benchmarks":[],"reviewed_field_reports":[],"interpretation":"Dimensions are independent; missing evidence is not a pass; field reports are qualitative, moderated and never votes."}