API & MCP docs

Claim schema

Agents submit claims as a JSON array. Only kind and aiClaim are required.

{
  "kind": "function" | "structure" | "vulnerability" | "behavior",
  "title": "Stack overflow in parse_header",
  "aiClaim": "memcpy into 64-byte stack buffer with attacker length",
  "address": "0x401a3c"           // optional, VA or symbol
}

MCP endpoint

Exposed at https://verire.lovable.app/mcp. Protected with OAuth — agents sign in as you and only see your data.

claude mcp add verire https://verire.lovable.app/mcp --transport http
# You'll be prompted to sign in (OAuth) on first use.

MCP tools

Three tools are available to connected agents.

list_binaries()                       -> your uploaded binaries
submit_claim(binary_id, claims[])       -> queued for verification
get_claim_verdict(binary_id)            -> status, method, checks, evidence

Verification response

Every claim gets deterministic checks plus an LLM analysis pass. A failed deterministic check caps the verdict at inconclusive.

{
  "status": "grounded" | "refuted" | "inconclusive",
  "method": "llm-analysis",
  "confidence": 0.87,
  "checks": [{ "name": "SHA-256 matches upload record", "passed": true }],
  "evidence": { "summary": "...", "strings": ["..."] }
}