Citare

Structured paper understanding for AI

Claims · causal metadata · integrity warnings — for honest citation

DOI is for papers. Citare is for claims.

—

Papers

—

Claims

—

Relations

—

Integrity

Citare is a database of claims extracted from peer-reviewed academic papers. Each claim carries causal-strength metadata (design basis, verification status) and integrity warnings (mediator, boundary condition, effect-disappears-under-control), so AI can cite findings honestly.

① Try it now — search the graph

Try:

Loading…

Claim graph: — depth:

Edge color: effect_disappears / replication / retracted hub_component boundary_condition extends_prior_definition none

② Connect to your AI

Click to show connection methods

Claude Code

One command. No install.

claude mcp add --transport http citare https://citare.dev/mcp

Claude.ai Web

Customize → Connectors → + → Add custom connector → URL:

https://citare.dev/mcp

Claude Desktop

Settings → Developer → Edit Config → add to mcpServers:

"citare": {"transport": "http", "url": "https://citare.dev/mcp"}

Other MCP clients (Codex, Gemini, ChatGPT, legacy SSE)

Codex

codex mcp add citare --url https://citare.dev/mcp --transport http

Gemini CLI

Add to ~/.gemini/settings.json:

{"mcpServers": {"citare": {"url": "https://citare.dev/mcp"}}}

ChatGPT Web

Settings → Connectors → Create (Developer Mode required) → URL:

https://citare.dev/mcp

Legacy SSE endpoint

Deprecated — kept for clients that haven't migrated yet. New users should use /mcp above.

https://citare.dev/sse

Tools

search_claims: Search by query, DOI, iv/dv, or template type. FTS5-backed.
cite_claim: Full citation bundle: source quote, page, integrity warnings, safe verbs derived from study design.
get_claim_graph: BFS the local neighbourhood of a claim — surface mediation, moderation, effect-disappears warnings.
get_extraction_prompt: The locked v0.13d production prompt + sub-agent invocation guide. Pass verbatim to a fresh context.
get_pdf_acquisition_guide: 7-stage PDF acquisition playbook: local search, OA publishers, Unpaywall, web search, Europe PMC, ask user.
register_claims: Submit an extraction JSON to grow the graph. Quality-gated; no auth needed.

③ Contribute — let's build the knowledge base for AI together

Anyone can grow Citare.

Extract a paper with the v0.13d prompt and call register_claims. No auth required. Pydantic schema and the quality gate reject broken extractions.

get_pdf_acquisition_guide() → obtain the PDF
get_extraction_prompt() → fetch the v0.13d prompt and run extraction in a sub-agent
register_claims(json_data=<extracted JSON>) → register to the DB
search_claims(doi=...) → verify the registration

Daily Dropbox backups + WARNING-not-REJECT design make it safe to experiment.

Resources

Strategy — mission, vision, and the commitments that follow
Interaction Design Document — how LLMs are designed to use Citare
Design Review for Builders — feedback for those building similar systems
GitHub: citareMCP — MCP server source
Live API stats (JSON)

Citare — AI-native scholarly claim infrastructure
MCP endpoint: https://citare.dev/mcp