About this project

academic-tools-mcp is a Model Context Protocol (MCP) server built on FastMCP that equips LLM agents with focused tools for working with academic papers. It supports unified paper metadata retrieval, author lookups, full PDF download and conversion to sectioned markdown, reference and citation graph traversal, and Wikipedia cross-referencing. Data sources include OpenAlex (paper metadata, authors, citations), arXiv and bioRxiv/medRxiv (preprints with PDF access), ACL Anthology (conference papers), Crossref (reference lists and DOI discovery), OpenCitations (citation graphs), and Wikipedia (article summaries). All API responses are cached locally with provider-specific TTLs, and concurrent calls for the same paper are coalesced into a single fetch. Definitive 404s are negative-cached to avoid burning rate budgets. The PDF pipeline downloads papers and converts them to section-level markdown using configurable backends such as MinerU or Marker, or a lightweight fast mode via pdftotext. A BM25 search index covers all converted papers for local full-text search. Setup requires Python 3.11+ and uv. Configuration is via environment variables in .env. The server can be used with Claude Code, run standalone, or launched via the FastMCP CLI.