Search Academic Papers from arXiv

mcp-scholarly is an MCP server focused on finding accurate academic articles. In its current form it implements a single tool, search-arxiv, which takes a required keyword string argument and searches arXiv for articles related to that keyword, returning results an AI assistant can cite or summarize. The project states more scholarly vendors will be added over time, so arXiv is the first of what is meant to be a broader academic-search surface.

Use this when an assistant needs to find real, citable academic papers on a topic - literature review, fact-checking a claim against published research, or surfacing recent preprints - rather than relying on an LLM's own possibly outdated or fabricated recollection of papers. It is not useful today for scholarly sources outside arXiv, since no other vendor integration is implemented yet, and it does not fetch full paper text or perform analysis - it returns search results for further use.

The single search-arxiv tool accepts one required parameter, keyword, and returns matching arXiv articles for that search term.

Installed via uv (either from source with a --directory pointing at the project, or via uvx mcp-scholarly for the published package) or via Docker (docker run --rm -i mcp/scholarly), and configured in Claude Desktop's claude_desktop_config.json under mcpServers. It can also be installed automatically via Smithery.

Researchers, students, and anyone building an AI assistant workflow that needs to ground answers in real arXiv papers rather than an LLM's unaided recall of academic literature. It is also useful during development or debugging of the server itself, since the project can be launched through the MCP Inspector for step-by-step request/response visibility, given that MCP servers communicate over stdio and are otherwise hard to observe directly.

npx -y @smithery/cli install mcp-scholarly --client claude