arXiv MCP Server — Search and Analyze Papers

Practical Notes

• README provides multiple install paths (uv tool install, Smithery, release artifacts) and documents --storage-path with a default papers directory.
• README includes stdio config snippets for uvx/uv tool run and mentions a portable .mcp.json for discovery.

Make Paper Search Reproducible

To avoid “I can’t find that paper again”:

• Always store the exact query terms and filters in your notes (arXiv categories, date ranges).
• Keep a dedicated --storage-path for PDFs/metadata so summaries and citations can link back to stored artifacts.
• Ask the agent to output a structured shortlist: title, authors, arXiv ID, published date, 3–5 key claims, and your “why it matters” note.

Suggested Agent Prompt Pattern

“Search arXiv for X, filter to the last N years, then give me 5 papers with arXiv IDs and a one-paragraph summary each. End with a reading order.”

FAQ

Q: Why not uv pip install? A: The README warns uv pip install may not put the executable on PATH; it recommends uv tool install.

Q: Can I customize storage? A: Yes—README documents --storage-path and a default papers directory.

Q: Is there a standard client config? A: Yes—README includes mcpServers JSON snippets for stdio launch via uvx/uv tool.