The rest of verification is lookup. A certified knowledge base has two characteristic weaknesses — a claim that was true when certified and is not now, and a statement no stored claim confirms but one contradicts — and addressing both directly is more thoughtful than a confidence score. get_my_usage reporting which claims you depend on is the same instinct applied downstream: if a claim you built on is later revised, you can find what it affected. The caveat is unavoidable — verification is only as good as the corpus, so list_domains and the claim counts are worth reading before treating a pass as meaningful. A statement outside the covered domains is unverified, not verified true.
A claim-verification service with three layers: searching and retrieving certified claims, verifying text against them, and an agent-account layer with a trust score and domain subscriptions.
- verify_response checks a full generated paragraph for hallucination risk, which is a different unit from verifying one claim and is the one that fits how text is actually produced.
- verify_claim and verify_claims_batch check single statements or many in parallel.
- find_contradictions looks for certified claims that contradict a statement — the inverse search, and the one that catches confident errors.
- get_citations returns formatted citations for a claim, and get_claim retrieves one with its full provenance.
- check_freshness asks whether a fact is still current, which is the failure mode a static knowledge base otherwise has.
- check_regulatory_compliance checks generated text against Swiss and EU regulatory requirements.
- get_knowledge_brief returns a structured citable brief on a topic, and list_domains shows the domains with their certified claim counts.
- submit_claim puts a new claim into expert review, with get_claim_status following it.
- An account layer: register the agent for a public trust score, read that score, subscribe to domain updates by webhook, read usage and which claims you depend on, and report what is missing.
Registration for the account-level features. The verification tools work against the service's own certified corpus, so coverage is whatever list_domains reports.
One command — npx -y mcp-remote https://swisstruth.org/mcp