MCP PDF Reader

Available Languages: 🇬🇧 English | 🇪🇸 Español

A powerful Model Context Protocol (MCP) server that empowers AI assistants like Claude and GitHub Copilot to intelligently interact with PDF documents. Extract text, metadata, search content, and retrieve embedded images—all through a standardized, LLM-friendly interface. Not OCR-based.

Current Version: 1.0.0
Package: @rturv/mcp-pdf-reader
License: MIT

Quick Start

Installation

Run the Server

Configuration

Claude Desktop

Add to claude_desktop_config.json:

Location:

• Windows: %APPDATA%\Claude\claude_desktop_config.json

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Linux: ~/.config/claude/claude_desktop_config.json

GitHub Copilot (VS Code)

Add to mcpServers.json:

Location: %APPDATA%\Code\User\globalStorage\github.copilot-chat\mcpServers.json

See COPILOT_CONFIG.md for additional installation methods.

Features

• ✅ Full Text Extraction - Extract complete text from PDF files

• ✅ Metadata Extraction - Retrieve title, author, creation date, and more

• ✅ Page Range Reading - Extract text from specific pages

• ✅ Text Search - Find text with surrounding context

• ✅ Page Count - Get total page count

• ✅ Image Extraction - List and extract embedded images in Base64

• ✅ Standards Compliant - Follows MCP specification for seamless LLM integration

Tools Reference

This MCP server exposes 7 tools for comprehensive PDF manipulation. All tools are accessible through Claude Desktop, GitHub Copilot, and other MCP-compatible clients.

1. read_pdf

Purpose: Extract all text content from a PDF file. Use this as your primary method for understanding PDF document content.

When to use:

• Reading entire document content

• Getting full document text for summarization or analysis

• Extracting content when combined with metadata

Input Parameters:

Example Request:

Example Response:

2. get_pdf_metadata

Purpose: Extract document metadata without reading the full text. Ideal for quick document inspection.

When to use:

• Identifying document properties (author, title, dates)

• Quick document validation

• Building document catalogs

• Checking modification dates

Input Parameters:

Example Request:

Example Response:

3. read_pdf_pages

Purpose: Extract text from a specific page or range of pages. Use this for targeted content extraction.

When to use:

• Reading specific sections of a document

• Analyzing particular chapters or pages

• Extracting cover pages or specific reports within a multi-part document

• Handling large PDFs by reading sections

Input Parameters:

Example Request (single page):

Example Request (page range):

Example Response:

4. search_pdf

Purpose: Search for text within a PDF and retrieve all matches with surrounding context.

When to use:

• Finding specific terms or phrases

• Locating sections by keyword

• Validating content presence

• Building keyword-based summaries

• Compliance checking (finding specific clauses)

Input Parameters:

Example Request:

Example Response:

5. get_pdf_page_count

Purpose: Get the total number of pages in a PDF without reading content.

When to use:

• Validating PDF integrity

• Determining if a PDF is empty

• Planning page range extractions

• Batch processing logic based on document size

Input Parameters:

Example Request:

Example Response:

6. list_pdf_images

Purpose: List all images embedded in a PDF with their metadata and locations.

When to use:

• Discovering embedded images before extraction

• Getting image dimensions and types

• Planning image extraction operations

• Validating image content presence

Input Parameters:

Example Request:

Example Response:

7. extract_pdf_image

Purpose: Extract a specific image from a PDF and return it as Base64-encoded data.

When to use:

• Recovering images from PDFs

• Converting PDF images to standard formats

• Processing visual content for analysis

• Archiving embedded images

Before using: Call list_pdf_images first to discover available images and their indices.

Input Parameters:

Example Request:

Example Response:

Note: Image data is Base64 encoded. Decode it to save as a file (e.g., using atob() in JavaScript or base64 -d in bash).

Testing with MCP Inspector

The MCP Inspector is an interactive tool for testing and debugging MCP servers. It provides a web-based UI to invoke tools and inspect responses in real-time.

Installation

Running the Server with Inspector

1. Start the MCP server in debug mode:

   mcp-pdf-reader

2. In a separate terminal, launch the Inspector:

   mcp-inspector node dist/index.js

   Or if using the global npm package:

   mcp-inspector npx @rturv/mcp-pdf-reader

3. Open the web UI: The Inspector will provide a URL (typically http://localhost:5173). Open it in your browser.

Using the Inspector

Example: Extract text from a PDF

1. In the Inspector UI, find the read_pdf tool in the left sidebar

2. Click on it to expand the tool interface

3. Enter the parameters:

   { "filePath": "C:/path/to/your/document.pdf", "includeMetadata": true }

4. Click "Call Tool"

5. View the response in the right panel

Example: Search for text

1. Select the search_pdf tool

2. Enter parameters:

   { "filePath": "C:/path/to/your/document.pdf", "searchTerm": "important keyword", "caseSensitive": false }

3. Click "Call Tool" and review the search results with context

Example: Extract an image

1. First, call list_pdf_images to discover images:

   { "filePath": "C:/path/to/your/document.pdf" }

2. Note the index of the image you want (e.g., index: 0)

3. Call extract_pdf_image with that index:

   { "filePath": "C:/path/to/your/document.pdf", "imageIndex": 0 }

4. The response will include Base64-encoded image data ready for decoding and saving

Troubleshooting Inspector Issues

• Port already in use: Change the port with mcp-inspector --port 5174

• Connection refused: Ensure the MCP server is running before starting the Inspector

• Tool not appearing: Verify the tool definition in src/index.ts and rebuild with npm run build

Development

Build

Watch Mode

Run Tests

Note: Tests require a sample PDF at test-files/sample.pdf. Create one or skip PDF-dependent tests.

Watch Tests

Project Structure

Technology Stack

• @modelcontextprotocol/sdk (^1.25.2) - MCP protocol implementation

• pdf-parse (^2.4.5) - PDF text extraction

• pdf-lib (^1.17.1) - PDF image extraction

• TypeScript (^5.9.3) - Type-safe development

• Jest (^29.7.0) - Unit testing

Limitations

• No OCR: Only extracts selectable text from PDFs (not scanned images)

• Text-based PDFs: Works best with PDFs containing embedded text. Scanned documents without OCR cannot be read

• Image extraction: Standard formats only (JPEG, PNG, TIFF)

• Base64 encoding: All images are returned as Base64 strings; large images may result in large responses

• No PDF modification: This server is read-only; it cannot edit or create PDFs

Configuration Files

• CLAUDE_CONFIG.md - Claude Desktop configuration guide

• COPILOT_CONFIG.md - GitHub Copilot setup instructions

License

MIT - See LICENSE file for details

Repository

github.com/rturv/mcp-pdf-reader