AI access — use the directory from an assistant

The directory can be queried by AI assistants through a Model Context Protocol (MCP) server. An assistant connected to it can search the 21399 investors, read a full record with its sources, and follow the graph — who shares portfolio companies with whom, and which limited partners disclose a commitment to a manager.

The endpoint

https://investordatabase.org/mcp — Streamable HTTP, read-only, no account or key.

Connecting

• Claude (claude.ai, desktop): Settings → Connectors → Add custom connector → paste the endpoint.
• Claude Code: claude mcp add --transport http investordatabase https://investordatabase.org/mcp
• ChatGPT and other clients that accept a remote MCP server: add the endpoint as a custom connector or MCP server URL.
• Your own code: any MCP client library, or plain JSON-RPC 2.0 over HTTP POST.

What it offers

• search_investors — free text, with filters for country, type, layer, stage, sector and city.
• get_investor — one record: type, headquarters, where and at what stage it invests, sectors, cheque size, its own thesis sentence, and the pages each fact was read from.
• get_co_investors — investors that list the same portfolio companies on their own pages.
• get_lp_relationships — limited partners that disclose a commitment to a manager, or the funds a limited partner discloses.
• directory_stats — what the directory covers, by type and country.

Limits, and why

At most 25 results a call, 200 per query, 60 requests a minute. The server is for answering questions, not for copying the directory; team members are on the record pages and are not served through it. Coverage is incomplete and uneven by country — an assistant should say so where it matters, and cite the record page it used.

A plain-text description of the site for language models is at /llms.txt. Questions or a broken answer: contact.