BGPT MCP

BGPT is an evidence layer for scientific AI. It gives AI assistants and developers structured records from research papers that link each claim to the experiments behind it, the exact results, scope, limitations, falsification criteria and source provenance, so an answer can be traced back to what a paper actually showed. The service is delivered as an MCP server and an HTTP API. Setup takes a configuration entry for Claude Desktop, Cursor or Claude Code, or a simple POST request from Python, using the SSE endpoint or the streamable HTTP endpoint. Two tools are exposed. Search papers takes a query with optional result count, a recent papers filter and an output format of evidence, full or legacy, and lookup paper retrieves a record by DOI. Each record includes a central claim, evidence units with claim type, evidence status and basis, experiment design, reported statistics and links to source sections. The vendor's site also carries a browsable evidence feed organized by topic and says its data is updated daily. It is built for literature review, RAG, research agents and scientific model training. Pricing is metered by result, where one result is one structured paper record. The free tier includes 50 results with no sign up or API key, and after that billing is $0.02 per result with unlimited searches. The site sits behind a Cloudflare challenge that can block automated requests and some networks, though it loads normally in a regular browser.