Discovery

• GET /llms.txt — machine-readable index: live coverage, surfaces, and canonical endpoints.
• GET /llms-full.txt — every platform with its verification_status and machine URL.

Content negotiation

Request a platform's canonical URL /platform/{slug} with Accept: application/json (or an agent User-Agent) to receive clean JSON; browsers get HTML at the same URL.

Machine endpoints

• GET /api/machine/platform/{slug} — token-efficient JSON + schema.org JSON-LD; each finding carries quote, surface, risk, confidence, verification_status, and provenance (snapshot SHA-256, Wayback, offsets).
• GET /api/machine/search?q={query}&surface={surface} — verified clause search.
• GET /api/coverage — live tracked vs verified counts.
• POST /api/verified/ask — grounded chat that answers only from cited verified findings and refuses beyond evidence.
• GET /api/verified/export/{slug} — GRC report mapped to NIST AI RMF + ISO/IEC 42001.
• POST /api/corrections/{slug} — trigger an autonomous re-run.

MCP server

POST /api/mcp — JSON-RPC 2.0 (Streamable HTTP). Tools:

• get_platform_risk — Get the verified risk rating and per-surface findings (with provenance) for one AI platform by slug.
• compare_platforms — Compare verified risk ratings across multiple AI platforms by slug.
• search_clauses — Full-text search verified verbatim clauses; optionally filter by surface. Returns quotes with provenance.
• check_training_use — Does this platform train on user content? Returns the verified training_use findings + provenance.

Every payload includes a disclaimer and per-finding provenance so the result can be independently re-verified against the public snapshot hash and Wayback URL.