The arXiv MCP Server is a local Model Context Protocol server that gives AI agents a proper working interface to academic literature. It has about 3,228 stars on GitHub under the Apache-2.0 license and was most recently active in 2026. Instead of an agent skimming abstracts through a search box, it can work through papers the way a researcher does, section by section, with citations handled correctly.
The server’s differentiator is depth: it reads papers in their original LaTeX source, so an agent can pull the outline of a paper and then read one precise section rather than ingesting a flattened PDF. It generates BibTeX entries directly from arXiv metadata, supports topic watches that keep an eye on new papers in your areas, and stores papers on your own disk, so your library stays local. Search is optional rather than the center of the design. The working loop is deliberately simple: paper ID, outline, one section, citations. Installation takes one command through uvx with no repository clone or Python environment setup, and ready-made configurations are provided for MCP clients such as Claude Code, Cursor, and VS Code, as well as one-click install links. Because it runs locally as a standard MCP server, it also composes naturally with the other agent tools researchers already use, from planning a review to drafting related-work sections with verifiable references.
The arXiv MCP Server is free and open source under the Apache-2.0 license.
It is for researchers, graduate students, and engineers who use AI assistants for literature reviews and want accurate, section-level paper reading with clean citations.