pdf-reader

pdf-reader-mcp is a production-ready MCP server that exposes a PDF processing toolkit for AI agents. It supports extracting text, images, and metadata from PDFs, while preserving document ordering using Y-coordinate-based layout. The server is designed for parallel processing and batch workloads, allowing multiple PDFs to be processed concurrently to maximize throughput. Typical usage involves starting the MCP server via npx and then issuing requests that specify sources (paths or URLs) and the desired output selections (text, images, metadata, page counts, etc.). The API is oriented around a single consolidated tool that can handle whole documents or specific page ranges, with support for absolute/relative paths and batch processing to scale across CPUs.

Once running, you can leverage the provided Quick Start examples in the README to build request payloads. You can request full text extraction, metadata, and page counts in one go, or tailor requests to extract only the pieces you need (e.g., only images and page-level metadata). The server emphasizes per-page error isolation, ensuring that failures on a single page do not derail the entire document processing workflow. This makes it suitable for enterprise-grade document ingestion pipelines where reliability and predictable performance are critical.

Prerequisites:

• Node.js installed (recommended LTS version)
• Internet access to fetch the MCP package from npm

Installation steps:

1. Install/run the MCP server via npx (no local installation required):

npx @sylphx/pdf-reader-mcp

2. (Optional) Install globally for repeated use:

npm install -g @sylphx/pdf-reader-mcp

3. If you prefer manual configuration in Claude/Windsurf/etc., you can reference the following example to add the server:

// Example: Claude Desktop / Windsurf / Windsurf-like config
{
  "mcpServers": {
    "pdf-reader": {
      "command": "npx",
      "args": ["@sylphx/pdf-reader-mcp"]
    }
  }
}

4. Validate installation by running a basic quick start request (as shown in the README) to ensure the MCP server responds and processes a sample PDF.

Tips and notes:

• The server is designed for parallel processing; you can batch multiple PDFs in a single request.
• Absolute vs relative paths are supported (requires v1.3.0+ for absolute path handling).
• When using in production, consider configuring environment variables if the MCP package exposes them for tuning performance or logging (check the package docs for specifics).
• If you encounter issues with path resolution, ensure the working directory of the hosting process is appropriate or prefer URL-based sources when possible.
• The npm package name to reference in configurations is @sylphx/pdf-reader-mcp; use the provided examples to wire it into Claude, Windsurf, Cline, Warp, or other MCP managers.