AgentOracle

What is AgentOracle?

AgentOracle is a factual-claim verification service designed for AI agents, compliance teams, content workflows, and developer platforms. It decomposes submitted text into atomic claims, checks each claim with four independent research models, assigns a confidence score and action recommendation, and returns a signed receipt that can be verified offline.

Responses are machine-readable and can classify claims as supported, refuted, or unverifiable. The service exposes HTTP endpoints, an MCP server for compatible AI clients, and open receipt-verification implementations. Its x402 integration supports HTTP-native payments rather than conventional API keys.

How to Use AgentOracle

1. Use the free preview endpoint or connect the AgentOracle MCP server to a supported client with npx agentoracle-mcp.
2. Submit text containing the claims that need verification.
3. Review the per-claim verdicts, evidence, confidence scores, and ACT, VERIFY, or REJECT recommendation.
4. Store or independently verify the returned JWS receipt against the published public keys when an audit trail is required.

Core Features

• Claim decomposition: Splits complex statements into independently verifiable atomic claims.
• Four-source verification: Runs four research models in parallel and uses agreement or contradiction to inform the result.
• Confidence and action output: Provides a 0.00–1.00 confidence score with an ACT, VERIFY, or REJECT recommendation.
• Signed verification receipts: Issues Ed25519 JWS receipts that can be checked offline against a public JWKS endpoint.
• Agent integrations: Provides an MCP server for Claude Desktop, Cursor, Windsurf, and other MCP-compatible clients.
• x402 payments: Supports per-query payments over Base, SKALE, and Stellar without a conventional API key.

Use Cases

• AI action gates: Agent developers verify a claim before allowing an automated system to take a consequential action.
• Content review: Brand and content teams fact-check AI-generated copy before publication and retain a signed audit trail.
• Compliance records: Compliance teams attach tamper-evident verification receipts to review workflows.
• Embedded verification: Developer-platform teams integrate claim verification as a feature in their own agent products.

Pricing

The /preview endpoint is free with truncated results and a limit of 20 requests per hour. /evaluate is free during beta. The website lists deep research at $0.10 per query and also identifies a research endpoint at $0.02. Paid endpoints have a stated limit of 100 requests per hour, while platform integrations use volume pricing.

Frequently Asked Questions

Does AgentOracle require an API key?

No. The service uses x402 payments for paid requests, and its free preview does not require a wallet or payment.

Can receipts be verified without AgentOracle?

Yes. The site provides a public key set, an open receipt format, and verifier implementations for offline checking.

What verdicts does AgentOracle return?

Each claim can be marked supported, refuted, or unverifiable, together with evidence, confidence, and an action recommendation.