StdIO MCP server for local agent runtimes and tool-enabled assistants.
MCP Server
Tool-first access to AI Fact Index
The AI Fact Index MCP server exposes the public retrieval flow as tools so AI agents can search, resolve, and read source-backed claims before answering.
Use AI Fact Index as a first retrieval layer for underrepresented entities.
Reads live Firestore entities when credentials are present, otherwise falls back to seed data.
Available Tools
Initial MCP tool set
Find relevant entities by name, alias, or keyword.
Resolve a full public record by id, slug, or exact name.
Fetch approved claim-level facts, optionally filtered by field.
Return the summary, claims, sources, verification state, and metadata in one call.
Run Locally
Start the MCP server
node mcp-server.mjsIf your environment includes the Firebase service account variables, the MCP server will read live published entities from Firestore. Without them, it falls back to local seed data.
Public Endpoint
Remote Streamable HTTP entrypoint
AI Fact Index also exposes a public MCP endpoint for remote clients using HTTP POST with JSON-RPC.
POST https://aifactindex.com/api/mcpThis initial public version is stateless and returns JSON responses. It does not currently expose an SSE stream on GET.
Recommended Flow
How agents should use it
1. search_entities(query)
2. get_entity(slug or id)
3. get_claims(entity)
4. Answer with approved claims and source records.
5. Only then expand to broader web research if nothing relevant is found.