Transparency Report — Live Data

Built for AI Agents
that need facts, not guesses.

Swiss Truth is a certified knowledge base with source-backed, expert-reviewed claims and SHA256 integrity hashes — ready to plug into your agent via MCP.

View MCP Discovery API Reference
Quick Setup
Add to Claude Desktop — one paste, done.
Paste into your config file and restart Claude Desktop.
claude_desktop_config.json
{ "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
✅ No API key required
By the Numbers

Live Knowledge Base Stats

All numbers are computed from the live database and updated on every page load.

2014
Certified claims
95.3%
Avg. confidence score
1470
Unique primary sources
28
Knowledge domains
Languages supported
Deutsch 1990 claims
English 24 claims
Methodology

5-Stage Validation Pipeline

Every claim passes all five stages before receiving a certified status and being made available to AI agents.

1
Submission & Deduplication
Incoming claims are embedded with a multilingual sentence transformer. If a semantically identical claim (≥ 95% cosine similarity) already exists, the submission is rejected to prevent redundancy.
Automated
2
AI Pre-Screen (Claude Haiku)
Each claim is evaluated by Claude Haiku for atomicity (one clear statement), factuality (not opinion or prediction), and source presence. Compound or opinion-based claims are flagged and returned as draft.
Automated
3
Source URL Verification
Each submitted source URL is fetched, and its content is checked by Claude Haiku to confirm it directly supports the claim. Wikipedia is excluded as a primary source. Only institutional, governmental, and peer-reviewed sources are accepted.
Automated
4
Expert Peer Review
Claims that pass automated screening enter the review queue for human expert validation. Reviewers assign a confidence score (0–1) and either certify or reject. A claim requires at least one expert approval to be certified.
Human review
5
SHA256 Integrity Signing & Expiry
Certified claims receive a SHA256 hash over their canonical content for tamper detection. Every claim has an expiry date (typically 12 months). Expired claims enter the renewal queue for re-validation — confidence scores decay linearly at 1% per month until renewed.
Cryptographic integrity
Coverage

Knowledge Domains

All certified claims by domain. Claims without a domain assignment are not included in the MCP search results.

Domain Certified Claims
AI/ML 112
Global Science 112
Weltgeschichte 109
EU Gesundheit 105
Weltraumwissenschaft 103
Naturwissenschaften 102
Klimawissenschaft 100
Schweizer Energie 93
Schweizer Verkehr 91
Schweizer Finanzmarkt 89
Schweizer Politik 87
Schweizer Landwirtschaft 85
Schweizer Bildung 81
Internationales Recht 81
Schweizer Gesundheitswesen 72
US-Recht & Regulierung 71
Weltwirtschaft 64
Cybersecurity 56
Quantencomputing 53
Biotechnologie 47
Erneuerbare Energien 47
Schweizer Digitalpolitik 45
Schweizer Recht 43
AI Safety 39
Schweizer Immobilienmarkt 5
AI/ML (English) 3
EU-Recht 2
EU Law & Regulation 1
Source Quality

Where claims come from

Primary sources must directly support the claim. Wikipedia is excluded. We prioritize governmental institutions, academic publications, and peer-reviewed research.

🏛️
Government & Official
546
admin.ch, parlament.ch, fedlex, cantonal authorities
🔬
Academic & Research
262
arxiv.org, nature.com, NIH, university publications
📰
Other Primary Sources
662
Industry standards, reports, and verified institutional sources
Integration

Connect your agent in 30 seconds

Swiss Truth speaks MCP — the Model Context Protocol standard for AI tool use. No API key required.

Available MCP Tools
TOOL
search_knowledge
Semantic search over certified claims in DE/EN/FR/IT/ES/ZH — returns confidence score, sources, and SHA256 hash
TOOL
verify_claim
Fact-check a statement → verdict: supported / contradicted / unknown with evidence and source references
TOOL
submit_claim
Submit a new claim for expert review — triggers the 5-stage validation pipeline automatically
TOOL
get_claim · list_domains · get_claim_status
Retrieve claim provenance, list all domains, or track submission status
MCP Auto-Discovery: The /.well-known/mcp.json endpoint follows RFC 8615 — compatible tools can discover and configure this server automatically.
Stay Updated

Subscribe to new certified claims

Two ways to get notified when a new claim is certified — pick what fits your stack.

📡
RSS Feed
Poll or subscribe with any RSS reader or agent. Returns the latest 50 certified claims as RSS 2.0 — no signup needed.
# All domains
https://swisstruth.org/feed.rss

# Filter by domain
https://swisstruth.org/feed.rss?domain=swiss-health
Open Feed →
🪝
Webhook (Push)
Register a URL and receive a POST within seconds of every certification. Optional domain filter. Returns an ID + token for unsubscribing.
# Subscribe
curl -X POST https://swisstruth.org/webhooks \
  -H "Content-Type: application/json" \
  -d '{"url":"https://your-agent.example.com/hook",
     "label":"my-agent",
     "domain":"swiss-health"}'

# Unsubscribe
curl -X DELETE \
  "https://swisstruth.org/webhooks/{id}?token={token}"
Webhook Payload
{
  "event":  "claim.certified",
  "source": "swiss-truth-mcp",
  "timestamp": "2026-04-10T12:00:00+00:00",
  "claim": {
    "id":               "01626b41-...",
    "text":             "In der Schweiz ist die Krankenversicherung...",
    "domain_id":        "swiss-health",
    "confidence_score": 0.99,
    "hash_sha256":      "sha256:d213...",
    "source_references": ["https://bag.admin.ch/..."],
    "language":         "de"
  },
  "mcp_endpoint": "https://swisstruth.org/mcp"
}