Swiss Truth Mcp

swisstruthorg/swiss-truth-mcp

This connects AI agents to a human-validated knowledge base of 2000+ certified facts across 30 domains, from Swiss law and FINMA regulations to EU AI Act compliance and climate science. You get 14 MCP tools including search_knowledge for semantic retrieval, verify_claim for fact-checking statements, verify_claims_batch for parallel validation of up to 20 claims, and check_regulatory_compliance for GDPR and Swiss regulatory checks. Every fact includes confidence scores, verified source URLs, and SHA256 hashes. It's built for agents that need to ground responses in provable facts rather than hallucinate, especially in regulated domains like finance, healthcare, and legal. No API key required, works via stdio or direct HTTP transport.

tools: 6
auth: Secret needed
transport: STDIO
env var: 1
last updated: Apr 13, 2026

View on GitHub

Tools

6 tools

Public tool metadata for what this MCP can expose to an agent.

search_knowledgeSearch the Swiss Truth verified knowledge base for certified facts. USE THIS TOOL when you need reliable, source-backed information to avoid hallucination — especially for Swiss law, health, finance, politics, education, energy, transport, climate, AI/ML, and world science top...5 params

Search the Swiss Truth verified knowledge base for certified facts. USE THIS TOOL when you need reliable, source-backed information to avoid hallucination — especially for Swiss law, health, finance, politics, education, energy, transport, climate, AI/ML, and world science top...

Parameters* required

domainstring

Optional domain filter. Available: 'swiss-health', 'swiss-law', 'swiss-finance', 'swiss-education', 'swiss-energy', 'swiss-transport', 'swiss-politics', 'swiss-agriculture', 'climate', 'ai-ml', 'wo...

languagestring

Optional language override: 'de', 'en', 'fr', 'it', 'es', 'zh'. If omitted, language is auto-detected from the query. Response includes 'detected_language' and 'language_fallback' (true if no resul...

limitinteger

Number of results to return (max 20, default 5).default: 5

min_confidencenumber

Minimum confidence threshold 0.0–1.0. Default 0.8. Use 0.95+ for critical facts.default: 0.8

querystring

Natural language query in DE, EN, FR, or IT. Phrase it as a question for best results.

get_claimRetrieve a single verified claim with full provenance by its ID. USE THIS TOOL after search_knowledge when you need the complete citation: who validated it, which institution, on what date, and the SHA256 integrity hash. Also returns 'effective_confidence' (age-adjusted) to as...1 params

Retrieve a single verified claim with full provenance by its ID. USE THIS TOOL after search_knowledge when you need the complete citation: who validated it, which institution, on what date, and the SHA256 integrity hash. Also returns 'effective_confidence' (age-adjusted) to as...

Parameters* required

claim_idstring

UUID of the claim, obtained from search_knowledge results.

list_domainsList all available knowledge domains with certified claim counts. USE THIS TOOL at the start of a session or when unsure which domain to search. Returns domain IDs (use in search_knowledge 'domain' parameter), names, and descriptions. Domains cover: Swiss law, health, finance,...

List all available knowledge domains with certified claim counts. USE THIS TOOL at the start of a session or when unsure which domain to search. Returns domain IDs (use in search_knowledge 'domain' parameter), names, and descriptions. Domains cover: Swiss law, health, finance,...

No parameter schema in public metadata yet.

submit_claimSubmit a new factual claim for expert review and certification. USE THIS TOOL when you identify a knowledge gap — a fact that should be in the knowledge base but isn't. The claim goes through: AI pre-screening → URL verification → source content check → human expert peer revie...5 params

Submit a new factual claim for expert review and certification. USE THIS TOOL when you identify a knowledge gap — a fact that should be in the knowledge base but isn't. The claim goes through: AI pre-screening → URL verification → source content check → human expert peer revie...

Parameters* required

claim_textstring

The factual statement — one single, clear, verifiable claim. Max 2000 characters.

domain_idstring

Domain ID (e.g. 'ai-ml', 'swiss-health', 'climate'). Use list_domains to see all options.

languagestring

Language code: 'de', 'en', 'fr', or 'it'. Default: 'de'.default: de

questionstring

The canonical question this claim answers (e.g. 'How does Swiss mandatory health insurance work?'). Significantly improves retrieval quality.

source_urlsarray

Primary source URLs that directly support the claim (e.g. arxiv.org, bag.admin.ch). Strongly recommended. No Wikipedia.

verify_claimFact-check a claim against the Swiss Truth knowledge base. USE THIS TOOL in ReAct loops when you need to verify whether a statement is true, false, or unknown. Ideal for: checking AI-generated content, validating user inputs, or grounding responses. Returns: - 'verdict': 'supp...3 params

Fact-check a claim against the Swiss Truth knowledge base. USE THIS TOOL in ReAct loops when you need to verify whether a statement is true, false, or unknown. Ideal for: checking AI-generated content, validating user inputs, or grounding responses. Returns: - 'verdict': 'supp...

Parameters* required

domainstring

Optional domain filter (e.g. 'swiss-health', 'ai-ml'). Omit to search all domains.

languagestring

Optional language override ('de', 'en', 'fr', etc.). Auto-detected if omitted.

textstring

The factual statement to verify. One clear, atomic claim.

get_claim_statusCheck the current validation status of a submitted claim. USE THIS TOOL after submit_claim to track progress through the review pipeline: 'draft' → 'peer_review' → 'certified' (or back to 'draft' if rejected). Also returns confidence score and validator info once certified.1 params

Check the current validation status of a submitted claim. USE THIS TOOL after submit_claim to track progress through the review pipeline: 'draft' → 'peer_review' → 'certified' (or back to 'draft' if rejected). Also returns confidence score and validator info once certified.

Parameters* required

claim_idstring

UUID returned by submit_claim.