ArXiv + Semantic Scholar Server

An MCP server that gives Claude (or any MCP client) the ability to search academic papers, explore citation graphs, and discover related research.

Tools

Quick Start

1. Install dependencies

# Using uv (recommended)
cd arxiv-scholar-mcp
uv venv
source .venv/bin/activate   # or .venv\Scripts\activate on Windows
uv pip install -e .

# Or using pip
pip install -e .

2. (Optional) Set a Semantic Scholar API key

Without a key, you share rate limits with all unauthenticated users. Request a free key at https://www.semanticscholar.org/product/api#api-key-form

export SEMANTIC_SCHOLAR_API_KEY="your-key-here"

3. Configure Claude Desktop

Add this to your claude_desktop_config.json:

Using uv (recommended):

{
  "mcpServers": {
    "arxiv-scholar": {
      "command": "uv",
      "args": [
        "--directory", "/ABSOLUTE/PATH/TO/arxiv-scholar-mcp",
        "run", "server.py"
      ],
      "env": {
        "SEMANTIC_SCHOLAR_API_KEY": "your-key-here"
      }
    }
  }
}

Using python directly:

{
  "mcpServers": {
    "arxiv-scholar": {
      "command": "/ABSOLUTE/PATH/TO/arxiv-scholar-mcp/.venv/bin/python",
      "args": ["/ABSOLUTE/PATH/TO/arxiv-scholar-mcp/server.py"],
      "env": {
        "SEMANTIC_SCHOLAR_API_KEY": "your-key-here"
      }
    }
  }
}

4. Restart Claude Desktop

The tools should now appear in your Claude Desktop conversation.

Example Prompts

• "Search ArXiv for recent papers on diffusion models for protein design"
• "Get details on ArXiv:2301.12345"
• "Who cites the 'Attention Is All You Need' paper?"
• "Find papers related to DOI:10.48550/arXiv.1706.03762"
• "Look up author Yann LeCun and show me his top papers"
• "What are the references in ArXiv:2303.08774?"

Paper ID Formats

The Semantic Scholar tools accept multiple identifier formats:

Testing

You can test the server directly using the MCP Inspector:

cd arxiv-scholar-mcp
npx @modelcontextprotocol/inspector uv run server.py

License

MIT

Thank you to arXiv for use of its open access interoperability. Data source: Semantic Scholar API (https://www.semanticscholar.org/product/api)