Swiss Truth MCP

Stop your AI agent from hallucinating facts. Swiss Truth is a verified, source-backed knowledge base accessible via MCP — certified claims with confidence scores, primary source URLs, and SHA256 integrity hashes.

No API key. No setup. Connect in 30 seconds.

---

Why agents need this

LLMs hallucinate facts, especially for country-specific, regulatory, and scientific topics. Swiss Truth gives your agent a ground-truth layer it can query before answering — returning only claims that have passed a 5-stage human + AI validation pipeline.

Without Swiss Truth	With Swiss Truth
"Health insurance in Switzerland is optional, I think..."	"Health insurance is mandatory (KVG Art. 3) — confidence 0.99, source: bag.admin.ch"
Unknown answer on Swiss VAT rates	Certified claim with current rate + legal source
Unverified AI/ML definition	Peer-reviewed explanation with academic citation

---

Quick Setup — Claude Desktop

{
  "mcpServers": {
    "swiss-truth": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://swisstruth.org/mcp"]
    }
  }
}

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

Restart Claude Desktop — done.

---

6 MCP Tools

search_knowledge — Semantic search over certified facts

search_knowledge("How does health insurance work in Switzerland?")
search_knowledge("Was ist RAG in der KI?")          # DE auto-detected
search_knowledge("三权分立原则")                       # ZH auto-detected

Returns: ranked list of certified claims, each with:

• confidence_score (0.0 – 1.0)
• source_urls (gov, academic, institutional — no Wikipedia)
• hash (SHA256 for tamper detection)
• language, domain, last_reviewed

Supports: DE · EN · FR · IT · ES · ZH · AR · RU · JA · KO

---

verify_claim — Fact-check any statement

verify_claim("Die Krankenversicherung in der Schweiz ist freiwillig.")

{
  "verdict": "contradicted",
  "confidence": 0.879,
  "explanation": "Swiss law (KVG Art. 3) mandates health insurance for all residents.",
  "evidence": [...],
  "sources": ["https://www.bag.admin.ch/..."]
}

Returns: supported · contradicted · unknown

---

get_claim — Full provenance for a single claim

get_claim("sha256:d21321db714d...")

Returns: claim text, domain, language, confidence, validator, institution, review date, all source URLs, SHA256 hash.

---

list_domains — Browse the knowledge base

list_domains()

Returns all 12 domains with certified claim counts. Use to understand coverage before querying.

---

submit_claim — Contribute to the knowledge base

Submit a claim for expert review. Automatically triggers the full validation pipeline (dedup → AI pre-screen → source verification → human review → signing).

---

get_claim_status — Track validation progress

Check where your submitted claim is: draft → peer_review → certified

---

Knowledge Domains

Domain	ID	Focus
Swiss Health	swiss-health	KVG, mandatory insurance, Krankenkasse
Swiss Law	swiss-law	Federal law, cantonal rules, civil code
Swiss Finance	swiss-finance	AHV, 3-pillar system, taxes, banking
Swiss Politics	swiss-politics	Federal Council, direct democracy, elections
Swiss Education	swiss-education	University system, apprenticeships, Bologna
Swiss Energy	swiss-energy	Nuclear, renewables, energy strategy 2050
Swiss Transport	swiss-transport	SBB, highways, LSVA
Swiss Agriculture	swiss-agriculture	Direct payments, organic, food sovereignty
Climate Science	climate	IPCC findings, temperature data, emissions
AI / ML	ai-ml	RAG, transformers, LLM concepts, benchmarks
Natural Sciences	world-science	Physics, chemistry, biology fundamentals
World History	world-history	Verified historical facts and dates

---

Agent Use Cases

RAG grounding — Before answering a user question, call search_knowledge to retrieve certified context. Pass it as system context to prevent hallucination.

Fact-checking pipeline — Use verify_claim as a post-processing step to validate claims in generated text before showing them to users.

Compliance checks — For Swiss regulatory topics (insurance, taxes, legal obligations), retrieve ground truth with source references your users can verify.

Multi-language support — One agent, 10+ languages. The query language is auto-detected — no need to specify it.

---

Trust & Methodology

Every claim passes a 5-stage validation pipeline:

1. Semantic dedup — vector similarity ≥ 95% blocks redundant submissions
2. AI pre-screen — Claude Haiku checks atomicity, factuality, source presence
3. Source verification — each URL fetched and verified to support the claim
4. Expert peer review — human validation with confidence score assignment
5. SHA256 signing + annual expiry — confidence decays 1%/month until renewed

Full transparency: swisstruth.org/trust

---

Endpoints

MCP endpoint	https://swisstruth.org/mcp
MCP discovery	https://swisstruth.org/.well-known/mcp.json
Trust & stats	https://swisstruth.org/trust
REST API docs	https://swisstruth.org/docs
RSS feed	https://swisstruth.org/feed.rss
Webhook subscriptions	POST https://swisstruth.org/webhooks

---

Subscribe to new certified claims

RSS — poll or subscribe in any feed reader:

https://swisstruth.org/feed.rss
https://swisstruth.org/feed.rss?domain=swiss-health

Webhook — get notified on every certification:

curl -X POST https://swisstruth.org/webhooks \
  -H "Content-Type: application/json" \
  -d '{"url": "https://your-agent.example.com/hook", "domain": "swiss-health"}'

---

Built with FastAPI · Neo4j · MCP · Claude