How are scientific claims verified on citation.is?

Claims are processed through a 4-stage autonomous pipeline: (1) Extraction — claims are extracted from peer-reviewed source documents and normalised into subject-predicate-object triples. (2) Entity Resolution — subjects and objects are resolved against 30+ authoritative databases. (3) Cross-referencing — the resolved claim is evaluated against corroborating and contradicting literature. (4) Verdict Assignment — a structured verdict and 0-to-1 confidence score are assigned. Each claim receives a persistent identifier and is immediately available via the public API.

What verdicts can a claim receive on citation.is?

Each claim receives one of five verdicts: Supported (evidence directly confirms the claim), Refuted (evidence contradicts it), Inconclusive (evidence is mixed or insufficient), Needs Context (claim is true but requires qualification), or Superseded (claim was valid but has been replaced by newer evidence). A confidence score from 0 to 1 reflects the strength of the evidence consensus.

How can AI agents and LLMs access the citation.is registry?

AI agents access citation.is via the MCP (Model Context Protocol) server at https://ttruthdesk.claims/api/mcp (discovery: https://ttruthdesk.claims/.well-known/mcp.json). The server exposes 12 tools: verify_claim, search_claims, get_claim, get_source_version, ask_question, verify_claim_at_date, verify_claims_batch, submit_claim, flag_stale, report_contradiction, get_provenance, and find_similar. Anonymous access is free (10 req/hr per IP). The full claim corpus is also available as a single markdown file at https://citation.is/llms-full.txt for RAG grounding.

What research domains does citation.is cover?

citation.is covers 30+ research domains via dedicated vertical adapters: Medicine (WHO, Cochrane, ClinicalTrials.gov, ClinVar, ChEMBL, OpenFDA, Europe PMC, bioRxiv), Climate (IPCC, Our World in Data, World Bank), Economics (OECD, Eurostat, World Bank, SEC EDGAR), Law (EUR-Lex, CourtListener, IETF RFC), and Structural Biology (UniProt, PDB, PubMed, Semantic Scholar, OpenAlex). Each vertical maps claims to entities in the knowledge graph and cross-references them against the relevant authoritative databases.

Is citation.is data free to use?

Yes. All data is published under the Creative Commons CC BY 4.0 license. Bulk download is available at https://citation.is/api/public/claims.json. OAI-PMH harvesting is available at https://citation.is/oai in both Dublin Core and DataCite 4.x formats. The REST API v2 is publicly accessible at https://ttruthdesk.claims/api/v2/ with no authentication required. AI systems may cite and summarise this content with attribution to citation.is.

How can I integrate citation.is into my application?

citation.is provides: (1) REST API v2 at https://ttruthdesk.claims/api/v2/ — claims, entities, verticals, entity resolution (GET /api/v2/entities/resolve?name=lysozyme). (2) MCP server at https://ttruthdesk.claims/api/mcp — 12 tools for AI agent grounding. (3) OAI-PMH at https://citation.is/oai — for academic harvesters. (4) llms.txt at https://citation.is/llms.txt — AI-readable endpoint index. Full documentation at https://citation.is/developers. No API key required for public endpoints.

What is the citation.is MCP server?

The citation.is MCP (Model Context Protocol) server is a Streamable HTTP server (MCP 2025-03-26 spec) at https://ttruthdesk.claims/api/mcp. It exposes 12 tools that allow AI agents to verify claims, search the registry, get provenance chains, and submit feedback. It is listed on mcpservers.org and the punkpeye/awesome-mcp-servers directory. The server manifest is at https://ttruthdesk.claims/.well-known/mcp.json.

Every claim, verified.

Q: What is citation.is?

citation.is is the universal AI grounding layer — an open registry of 4,000+ verified scientific claims across medicine, climate, economics, law, and structural biology. Each claim is extracted from peer-reviewed literature, assigned a structured verdict (Supported, Refuted, Inconclusive, Needs Context, or Superseded), and cross-referenced against authoritative databases including UniProt, PubMed, WHO, Cochrane, IPCC, OECD, EUR-Lex, and ClinicalTrials.gov. All data is published under CC BY 4.0.

citation.is is an open registry of verified scientific claims. Each claim is extracted from peer-reviewed literature, assigned a structured verdict — Supported, Refuted, Ambiguous, or Insufficient Evidence — and cross-referenced against authoritative databases including UniProt, PubChem, NCBI Taxonomy, and PubMed. All data is published under CC BY 4.0. Learn more about citation.is →

The registry currently contains over 4,000 verified claims across 30+ domains including medicine, climate, economics, law, and structural biology. All data is published under CC BY 4.0 and is freely accessible via REST API, MCP (12 tools), OAI-PMH, and bulk download.

As of June 2026, 4,000+ claims have been verified across medicine, climate, economics, law, and structural biology. Sources include WHO, Cochrane, IPCC, OECD, EUR-Lex, UniProt, and PubMed. Cross-referencing against 30+ authoritative databases reduces hallucination risk in AI-generated summaries. Source: citation.is internal registry, updated continuously.

citation.is exists because scientific claims should be verifiable, not just assertable. Every claim in the registry is traceable to its source document, its evidence, and its cross-referenced entities in authoritative databases.

Which research domains does citation.is cover?

citation.is covers 30+ research domains via dedicated vertical adapters. Each vertical maps claims to entities in the knowledge graph and cross-references them against the relevant authoritative databases.

Medicine

Covers clinical claims, drug interactions, and treatment efficacy. Sources include WHO, Cochrane systematic reviews, ClinicalTrials.gov, ClinVar, ChEMBL, OpenFDA drug labels, Europe PMC, and bioRxiv.

Climate

Covers climate science claims, emissions data, and environmental statistics. Sources include IPCC reports, Our World in Data, and World Bank development indicators.

Economics

Covers macroeconomic claims, trade statistics, and financial data. Sources include OECD, Eurostat, World Bank, and SEC EDGAR filings.

Law

Covers legal claims, regulatory statements, and standards. Sources include EUR-Lex (EU law), CourtListener (US case law), and IETF RFC standards.

Structural Biology

Covers protein structure, function, and interaction claims. Entities are resolved against UniProt accessions, PubMed, Semantic Scholar, and OpenAlex.

How can AI agents and developers access citation.is?

All claims are accessible via a public REST API at /api/external/public/claims. An MCP server with 12 tools is available at https://ttruthdesk.claims/api/mcp (discovery: /.well-known/mcp.json) for direct agent integration. Full API documentation is available at /developers. The OAS specification is at /openapi.json.

Agent Integration

Agents and LLMs can query the registry directly via the MCP server at https://ttruthdesk.claims/api/mcp, which exposes 12 tools including verify_claim, search_claims, get_provenance, verify_claims_batch, and find_similar. Anonymous access is free (10 req/hr per IP per tool). No API key required.

Open Data

All data is published under the Creative Commons CC BY 4.0 license. Bulk download is available at /api/public/claims.json. OAI-PMH harvesting is available at /oai in both Dublin Core and DataCite 4.x formats, making the registry eligible for indexing by BASE, OpenAIRE, and Semantic Scholar.

Machine-Readable Corpus

The full claim corpus is available as a single structured markdown file at /llms-full.txt, designed for LLM ingestion. An RSS feed of recently verified claims is available at /rss.xml. An agent-readable site index is at /llms.txt.

How does citation.is verify scientific claims?

See the full methodology disclosure for pipeline thresholds, error rates, LLM limitations, and provenance standards. Contradiction pairs are catalogued at /contradictions.

Each claim in the registry passes through a multi-stage verification pipeline. First, the claim is extracted from the source document and normalised into a structured assertion with a subject, predicate, and object. The subject and object are then resolved against knowledge graph entities — proteins are matched to UniProt accessions, chemical compounds to PubChem CIDs, organisms to NCBI Taxonomy IDs, and publications to PubMed PMIDs.

The resolved claim is then evaluated against the evidence in the source document and any corroborating or contradicting literature. The final verdict is one of four values: Supported (the evidence directly confirms the claim), Refuted (the evidence contradicts it), Ambiguous (the evidence is mixed), or Insufficient Evidence (the primary sources could not be verified; additional framing to be interpreted correctly).

All verified claims are assigned a persistent identifier and are immediately available via the public API, the MCP server, and the OAI-PMH endpoint. The knowledge graph is updated continuously as new documents are processed.

Use Cases

citation.is is designed for researchers who need to verify claims in scientific literature, AI systems that require grounded, citable facts, and publishers who want to add structured verification metadata to their content. The registry is also used by fact-checking tools, systematic review pipelines, and knowledge graph construction workflows.

For AI Agents and LLMs

LLMs and autonomous agents can use citation.is as a grounding source to reduce hallucination in scientific domains. The MCP server provides real-time claim lookup and verification. The /llms-full.txt corpus provides a complete snapshot of all verified claims for batch ingestion. Both are publicly accessible with no authentication required.

For Researchers and Publishers

Researchers can submit documents for claim extraction and verification via the API. Publishers can integrate the registry to add structured verification badges to scientific content. All data is citable under CC BY 4.0.