arxiv-mcp-server — Claude Code MCP server: features & setup
Connects your agent to arXiv papers to inspect original LaTeX sections, export BibTeX, and track topics locally without third-party cloud sync.
claude mcp add --env ARXIV_STORAGE_PATH=<ARXIV_STORAGE_PATH> --transport stdio arxiv-mcp-server -- uvx arxiv-mcp-serverWhat arxiv-mcp-server does
The server brings academic literature workflows directly to your local workstation. It lets the agent download papers and inspect author-submitted LaTeX section by section. The working loop starts from an arXiv ID, retrieves the outline, reads a specific section, and extracts BibTeX citations from authoritative metadata.
All papers remain on local disk under ~/.arxiv-mcp-server/papers. You can override this folder with --storage-path or the ARXIV_STORAGE_PATH environment variable. Running the server requires Astral's uv tool to execute uvx.
Do not install this server using npm, pnpm, or npx arxiv-mcp-server. An unrelated package occupies that name on the npm registry, so installation must run through PyPI via uvx.
What it helps with
- read specific arXiv paper sections directly from raw LaTeX
- extract accurate BibTeX citations from official metadata
- cache downloaded papers on local storage
- monitor targeted research topics for new preprints
Example prompts for the agent
- Fetch arXiv paper 2312.00752 and read the architecture section.
- Export the BibTeX citation for arXiv article 1706.03762.
- Check the outline of paper 2401.12345 and show the methodology.
How to install arxiv-mcp-server
Platform CLI
claude mcp add --env ARXIV_STORAGE_PATH=<ARXIV_STORAGE_PATH> --transport stdio arxiv-mcp-server -- uvx arxiv-mcp-serverReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from arxiv-mcp-server <command>
Platform CLI
codex mcp add arxiv-mcp-server --env ARXIV_STORAGE_PATH=<ARXIV_STORAGE_PATH> -- uvx arxiv-mcp-serverReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from arxiv-mcp-server <command>
Config file
{
"mcpServers": {
"arxiv-mcp-server": {
"command": "uvx",
"args": [
"arxiv-mcp-server"
],
"env": {
"ARXIV_STORAGE_PATH": "<ARXIV_STORAGE_PATH>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from arxiv-mcp-server <command>
Config file
{
"servers": {
"arxiv-mcp-server": {
"type": "stdio",
"command": "uvx",
"args": [
"arxiv-mcp-server"
],
"env": {
"ARXIV_STORAGE_PATH": "<ARXIV_STORAGE_PATH>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from arxiv-mcp-server <command>
Platform CLI
gemini mcp add -e ARXIV_STORAGE_PATH=<ARXIV_STORAGE_PATH> arxiv-mcp-server uvx arxiv-mcp-serverReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from arxiv-mcp-server <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about arxiv-mcp-server
How to install arxiv-mcp-server in Claude Code?
Run claude mcp add arxiv-mcp-server -- uvx arxiv-mcp-server in your terminal. To specify a custom storage directory, run claude mcp add --env ARXIV_STORAGE_PATH=<ARXIV_STORAGE_PATH> --transport stdio arxiv-mcp-server -- uvx arxiv-mcp-server.
What does arxiv-mcp-server do?
It provides local access to arXiv preprints, enabling agents to parse LaTeX sections, export BibTeX entries, and follow citation graphs.
Why should I avoid installing arxiv-mcp-server with npm?
An unrelated package uses the exact same name on npm. You must use uvx arxiv-mcp-server from PyPI to avoid downloading the wrong package.
