BGPT MCP

Searches scientific papers by indexing and querying full-text experimental methodology, results, and data sections rather than abstracts or titles. The system parses paper PDFs to extract experimental protocols, datasets, and findings, then applies semantic or keyword matching to surface papers based on methodological similarity or specific experimental approaches. This enables discovery of papers that traditional abstract-based search engines miss because the experimental details are buried in methods sections.

Exposes the paper search capability as a Model Context Protocol (MCP) server, allowing LLM agents and custom applications to call search functions directly within their tool-use workflows. The MCP integration handles request serialization, response formatting, and context passing between the client (Claude, custom agents) and the hosted search backend, enabling researchers to embed paper discovery into multi-step research automation pipelines without managing HTTP calls or authentication.

Provides 50 free searches without requiring account creation, API key registration, or authentication. The system likely uses IP-based or session-based quota tracking to enforce the 50-search limit per user, allowing immediate access for casual researchers and students without onboarding friction. This is implemented as a hosted service with no client-side authentication, making it accessible from any MCP-compatible client or web interface.

Parses scientific paper PDFs to extract and index experimental methodology, protocols, datasets, results, and findings at a granular level beyond abstracts. The system likely uses PDF text extraction, section detection (via heuristics or ML), and possibly named entity recognition to identify experimental parameters, measurements, and procedures. These extracted sections are then indexed in a searchable database, enabling queries that match on methodological similarity rather than keyword overlap.

Ranks search results based on semantic similarity between the user's query and extracted experimental data sections, rather than simple keyword matching or citation counts. The system likely uses embeddings (vector representations of text) to compare the user's methodological description with indexed experimental sections, returning papers where the experimental approach most closely matches the query intent. This enables finding papers with similar methodologies even if they use different terminology.