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/

The problem

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.

Tools

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

Standardized verdict schema

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": "sha256:a1b2c3...",
  "provenance": [...],
  "limitations": ["..."],
  "recommended_action": "caution"
}

Four action levels

Action	When	What agent should do
proceed	Confidence ≥ 85%	Evidence supports this action
caution	60–84%	Add uncertainty disclosure
abstain	30–59%	Do not make this claim
escalate	< 30%	Human review required

Example: Hallucination catch

{
  "name": "hallucination_check",
  "arguments": {
    "statement": "USDT is guaranteed risk-free and fully authorized under MiCA"
  }
}

→ Result: status: unverified | confidence: 25% | action: escalate

• Red flag: Absolute language detected ("guaranteed", "risk-free")
• Red flag: Factually suspect claim (USDT is NOT authorized under MiCA)

Infrastructure

The Trust Layer is backed by FeedOracle's Enterprise Trust infrastructure:

• ES256K signing — Every response cryptographically signed (RFC 7515)
• Evidence Registry — Append-only transparency log (CT-inspired)
• KYA (Know Your Agent) — Agent identity and trust scoring (0–100, 4 levels)
• On-chain anchoring — Polygon + XRPL for tamper-proof timestamping
• JWKS public — https://feedoracle.io/.well-known/jwks.json

Connect

{
  "mcpServers": {
    "trustlayer": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.feedoracle.io/trust/mcp/"]
    }
  }
}

Part of FeedOracle

The Trust Layer is part of FeedOracle compliance evidence infrastructure and powers the Decision Preflight monetization layer.

Part of the ToolOracle platform — 50+ MCP servers, 530+ tools.

License

MIT