trustoracle

Verifiable evidence & verification MCP server for AI agents.

Adds a verifiable trust layer so claims can be checked for provenance, freshness, and cryptographic integrity before action.

npx -y mcp-remote https://mcp.feedoracle.io/trust/mcp/

AI agents produce plausible-sounding outputs. But plausible is not verified. When agents make decisions about regulated assets, financial recommendations, or compliance determinations, ungrounded outputs create real risk.

The Trust Layer gives agents a verification gate: before acting, check the evidence.

| Tool | What it does | |------|-------------| | verify_claim | Check an agent's claim against FeedOracle signed evidence | | get_signed_fact | Get an ES256K-signed, machine-verifiable fact (JWKS public) | | provenance_trace | Full data chain: external source → FeedOracle → agent delivery | | freshness_guard | Is this data current, stale, or expired? Configurable max age | | hallucination_check | Compare statement against ground truth + red-flag pattern scan | | verify_signature | Verify JWS/DSSE signature from FeedOracle evidence | | confidence_gate | Decision gate: proceed / caution / abstain / escalate | | evidence_anchor | Anchor a fact on-chain (Polygon) as immutable timestamped proof | | agent_reputation | Trust score for an AI agent from KYA (Know Your Agent) system | | trust_status | Infrastructure health and capabilities |

Every tool returns this structure:

{
  "status": "partially_verified",
  "confidence": 65,
  "freshness": "current",
  "verification": {
    "signature_valid": true,
    "hash_match": true,
    "registry_entry": true,
    "source_count": 3
  },
  "sources": ["FeedOracle Stablecoin Feed", "CoinGecko", "DeFiLlama"],
  "signature": {
    "algorithm": "ES256K",
    "kid": "feedoracle-mcp-es256k-1",
    "jwks_url": "https://feedoracle.io/.well-known/jwks.json"
  },
  "hash"