PDF Reader MCP Server - MCP Server

PDF Reader MCP Server (@sylphlab/pdf-reader-mcp)

CI/CD Pipeline

codecov

npm version

Docker Pulls

License: MIT

Empower your AI agents (like Cline) with the ability to securely read and extract information (text, metadata, page count) from PDF files within your project context using a single, flexible tool.

Installation

Using npm (Recommended)

Install as a dependency in your MCP host environment or project:

pnpm add @sylphlab/pdf-reader-mcp # Or npm install / yarn add

Configure your MCP host (e.g., mcp_settings.json) to use npx:

{
  "mcpServers": {
    "pdf-reader-mcp": {
      "command": "npx",
      "args": ["@sylphlab/pdf-reader-mcp"],
      "name": "PDF Reader (npx)"
    }
  }
}

(Ensure the host sets the correct cwd for the target project)

Using Docker

Pull the image:

docker pull sylphlab/pdf-reader-mcp:latest

Configure your MCP host to run the container, mounting your project directory to /app:

{
  "mcpServers": {
    "pdf-reader-mcp": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-v",
        "/path/to/your/project:/app", // Or use "$PWD:/app", "%CD%:/app", etc.
        "sylphlab/pdf-reader-mcp:latest"
      ],
      "name": "PDF Reader (Docker)"
    }
  }
}

Local Build (For Development)

Clone: git clone https://github.com/sylphlab/pdf-reader-mcp.git
Install: cd pdf-reader-mcp && pnpm install
Build: pnpm run build

Configure MCP Host:

{
  "mcpServers": {
    "pdf-reader-mcp": {
      "command": "node",
      "args": ["/path/to/cloned/repo/pdf-reader-mcp/build/index.js"],
      "name": "PDF Reader (Local Build)"
    }
  }
}

(Ensure the host sets the correct cwd for the target project)

Quick Start

Assuming the server is running and configured in your MCP host:

MCP Request (Get metadata and page 2 text from a local PDF):

{
  "tool_name": "read_pdf",
  "arguments": {
    "sources": [
      {
        "path": "./documents/my_report.pdf",
        "pages": [2]
      }
    ],
    "include_metadata": true,
    "include_page_count": false, // Default is true, explicitly false here
    "include_full_text": false // Ignored because 'pages' is specified
  }
}

Expected Response Snippet:

{
  "results": [
    {
      "source": "./documents/my_report.pdf",
      "success": true,
      "data": {
        "page_texts": [
          { "page": 2, "text": "Text content from page 2..." }
        ],
        "info": { ... },
        "metadata": { ... }
        // num_pages not included as requested
      }
    }
  ]
}

Why Choose This Project?

🛡️ Secure: Confines file access strictly to the project root directory.
🌐 Flexible: Handles both local relative paths and public URLs.
🧩 Consolidated: A single read_pdf tool serves multiple extraction needs (full text, specific pages, metadata, page count).
⚙️ Structured Output: Returns data in a predictable JSON format, easy for agents to parse.
🚀 Easy Integration: Designed for seamless use within MCP environments via npx or Docker.
✅ Robust: Uses pdfjs-dist for reliable parsing and Zod for input validation.

Performance Advantages

Initial benchmarks using Vitest on a sample PDF show efficient handling of various operations:

| Scenario | Operations per Second (hz) | Relative Speed | | :------------------------------- | :------------------------- | :------------- | | Handle Non-Existent File | ~12,933 | Fastest | | Get Full Text | ~5,575 | | | Get Specific Page (Page 1) | ~5,329 | | | Get Specific Pages (Pages 1 & 2) | ~5,242 | | | Get Metadata & Page Count | ~4,912 | Slowest |

(Higher hz indicates better performance. Results may vary based on PDF complexity and environment.)

See the Performance Documentation for more details and future plans.

Features

Read full text content from PDF files.
Read text content from specific pages or page ranges.
Read PDF metadata (author, title, creation date, etc.).
Get the total page count of a PDF.
Process multiple PDF sources (local paths or URLs) in a single request.
Securely operates within the defined project root.
Provides structured JSON output via MCP.
Available via npm and Docker Hub.

Design Philosophy

The server prioritizes security through context confinement, efficiency via structured data transfer, and simplicity for easy integration into AI agent workflows. It aims for minimal dependencies, relying on the robust pdfjs-dist library.

See the full Design Philosophy documentation.

Comparison with Other Solutions

Compared to direct file access (often infeasible) or generic filesystem tools, this server offers PDF-specific parsing capabilities. Unlike external CLI tools (e.g., pdftotext), it provides a secure, integrated MCP interface with structured output, enhancing reliability and ease of use for AI agents.

See the full Comparison documentation.

Future Plans (Roadmap)

Documentation:
- Finalize all documentation sections (Guide, API, Design, Comparison).
- Resolve TypeDoc issue and generate API documentation.
- Add more examples and advanced usage patterns.
- Implement PWA support and mobile optimization for the docs site.
- Add share buttons and growth metrics to the docs site.
Benchmarking:
- Conduct comprehensive benchmarks with diverse PDF files (size, complexity).
- Measure memory usage.
- Compare URL vs. local file performance.
Core Functionality:
- Explore potential optimizations for very large PDF files.
- Investigate options for extracting images or annotations (longer term).
Testing:
- Increase test coverage towards 100% where practical.
- Add runtime tests once feasible.

Documentation

For detailed usage, API reference, and guides, please visit the Full Documentation Website (Link to be updated upon deployment).

Community & Support

Found a bug or have a feature request? Please open an issue on GitHub Issues.
Want to contribute? We welcome contributions! Please see CONTRIBUTING.md.
Star & Watch: If you find this project useful, please consider starring ⭐ and watching 👀 the repository on GitHub to show your support and stay updated!

License

This project is licensed under the MIT License.