arXiv Research Paper Analyzer MCP Server

An MCP (Model Context Protocol) server that enables LLM agents to search arXiv, download papers, parse PDFs into structured sections, and extract key findings using client-side LLMs. It features persistent local caching and layout-aware PDF extraction to optimize token usage and processing speed.

Architecture

The following diagram illustrates the workflow and architecture of the arXiv Analyzer server:

graph TD
    Agent([LLM Agent]) -->|Call Tool / Read Resource| Server[arXiv Analyzer MCP Server]
    Server -->|Query Meta / Download PDF| ArxivAPI[arXiv API]
    Server -->|Read Cached Files| Cache[Local File Cache]
    Server -->|Parse PDF Layout| MuPDF[PyMuPDF Parser]
    MuPDF -->|Segment Text| Sections[Parsed Sections]
    Server -.->|Request LLM completion| Sampling[Client LLM Sampling]
    Sampling -.->|Return Summary| Server

Related MCP server: paperstack

Project Structure

mcp-arxiv-analyzer/
├── .github/workflows/
│   └── ci.yml               # GitHub Actions CI pipeline
├── mcp_arxiv_analyzer/
│   ├── __init__.py
│   ├── arxiv_client.py      # arXiv API integration and downloader
│   ├── cache.py             # File caching layer for PDFs, sections, and summaries
│   ├── llm.py               # Client sampling interfaces and fallback logic
│   ├── pdf_parser.py        # PyMuPDF parser for layout-aware section segmentation
│   └── server.py            # MCP server endpoint and tool handlers
├── tests/                   # Unit and integration test suite
├── pyproject.toml           # Package configuration and dependencies
└── README.md                # Documentation

Tools Exposed

• search_arxiv: Search arXiv for research papers matching a query.

• download_paper: Download a paper by ID, parse its PDF into layout sections, and cache it.

• get_paper_sections: Get cached section names and outlines.

• get_section_text: Retrieve the raw text content of a specific section of a paper.

• get_paper_context: Retrieve a token-efficient pre-assembled overview context of the paper (Abstract, Intro, Conclusion, Outline, and focus-related sections) for summarization.

• extract_key_findings: Generate a structured summary of contributions, methods, experiments, and limitations. Falls back to providing context for client-side summarization if client-side sampling (CreateMessage) is unsupported.

Setup and Installation

Prerequisites

• Python 3.10 or higher

• uv package manager (recommended) or pip

1. Clone the Repository

git clone https://github.com/arman1o1/mcp-arxiv-analyzer.git
cd mcp-arxiv-analyzer

2. Setup Virtual Environment and Install

Using uv:

uv venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
uv pip install -e .

Using standard pip:

python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
pip install -e .

Running the Server

To run the MCP server on stdio (the standard protocol for MCP integrations):

python -m mcp_arxiv_analyzer.server

Configuration with MCP Hosts

To configure the server with your MCP host (like Claude Desktop or Antigravity), add the following server configuration to your mcp_config.json:

{
  "mcpServers": {
    "mcp-arxiv-analyzer": {
      "command": "uv",
      "args": [
        "--directory",
        "/absolute/path/to/mcp-arxiv-analyzer",
        "run",
        "python",
        "-m",
        "mcp_arxiv_analyzer.server"
      ]
    }
  }
}

Running Tests

To run the unit and integration tests locally, install development dependencies and run pytest:

pip install -e .[dev]
pytest