An MCP server that allows interaction with the arXiv API in natural language for searching, retrieving metadata, and downloading scientific papers.

Installation

From Source (MacOS)

git clone https://github.com/prashalruchiranga/arxiv-mcp-server.git
cd arxiv-mcp-server
brew install uv
uv venv --python=python3.13
source .venv/bin/activate
uv sync

From Source (Windows)

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
git clone https://github.com/prashalruchiranga/arxiv-mcp-server.git
cd arxiv-mcp-server
uv venv --python=python3.13
source .venv\Scripts\activate
uv sync

Configuration

Claude Desktop

{
  "mcpServers": {
    "arxiv-server": {
      "command": "uv",
      "args": [
        "--directory",
        "/ABSOLUTE/PATH/TO/PARENT/FOLDER/arxiv-mcp-server/src/arxiv_server",
        "run",
        "server.py"
      ],
      "env": {
        "DOWNLOAD_PATH": "/ABSOLUTE/PATH/TO/DOWNLOADS/FOLDER"
      }
    }
  }
}

Available Tools

Tool	Description
get_article_url	Get the URL of an article hosted on arXiv.org by its title
download_article	Download an article from arXiv.org in PDF format
load_article_to_context	Load an article from arXiv.org into the language model context
get_details	Get article metadata from arXiv.org by its title
search_arxiv	Perform a search query through the arXiv API with specified parameters and retrieve matching papers

Capabilities

• Retrieve metadata about scientific papers hosted on arXiv.org
• Download papers in PDF format to your local machine
• Search the arXiv database by specific query
• Retrieve papers and load them into a large language model (LLM) context

Environment Variables

Required

• DOWNLOAD_PATH - Absolute path to the downloads folder where PDF files will be saved

Usage Examples

Can you get the details of 'Reasoning to Learn from Latent Thoughts' paper?

Get the papers authored or co-authored by Yann Lecun on convolutional neural networks

Download the attention is all you need paper

Can you get the papers by Andrew NG which have 'convolutional neural networks' in title?

Can you display the paper?

Notes

Requires Python 3.13+ and the uv package manager. The configuration may require the full path to the uv executable (check using 'which uv' on MacOS or 'where uv' on Windows).