document-reader-mcp

Universal MCP server for extracting text from various document formats. Supports streaming, page/row limits, encoding detection, and simple rate limiting.

Cross-platform compatible: Works seamlessly on macOS, Linux, and Windows with identical functionality.

Supported Formats

Related MCP server: doc-ops-mcp

Features

✅ Cross-platform: Works on macOS, Linux, and Windows
✅ Multiple format support: PDF, Excel, CSV, TXT, JSON, Markdown, DOCX, PowerPoint, HTML
✅ Markdown conversion: Convert documents to Markdown with automatic image extraction
✅ PDF image extraction: Automatically extracts and embeds images from PDFs at appropriate page positions
✅ Streaming API: Memory-efficient processing of large files
✅ Smart encoding detection: Handles UTF-8, Latin-1, CP1252, ISO-8859-1
✅ Context-aware limits: Automatic truncation to prevent AI context overflow
✅ Rate limiting: Process-wide rate limiting (configurable)
✅ Docker support: Run in isolated container with non-root user
✅ Modular design: Easy to extend with new formats
✅ Minimal dependencies: Most formats use Python stdlib only

Installation

Option 1: Install from GitHub (Recommended)

macOS/Linux

# Clone the repository
git clone https://github.com/ifmelate/document-reader-mcp.git
cd document-reader-mcp

# Create virtual environment and install dependencies
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Windows (Command Prompt)

# Clone the repository
git clone https://github.com/ifmelate/document-reader-mcp.git
cd document-reader-mcp

# Create virtual environment and install dependencies
python -m venv .venv
.venv\Scripts\activate.bat
pip install -r requirements.txt

Windows (PowerShell)

# Clone the repository
git clone https://github.com/ifmelate/document-reader-mcp.git
cd document-reader-mcp

# Create virtual environment and install dependencies
python -m venv .venv
.venv\Scripts\Activate.ps1
pip install -r requirements.txt

Note for Windows PowerShell users: If you encounter an execution policy error, run:

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

Quick Setup Scripts

For convenience, you can use the provided setup scripts:

macOS/Linux:

chmod +x dev-setup.sh
./dev-setup.sh

Windows (Command Prompt):

dev-setup.bat

Windows (PowerShell):

.\dev-setup.ps1

These scripts will create the virtual environment, install dependencies, and set up the development environment automatically.

Option 2: Direct Install with pip

pip install git+https://github.com/ifmelate/document-reader-mcp.git

Option 3: Docker

# Clone the repository
git clone https://github.com/ifmelate/document-reader-mcp.git
cd document-reader-mcp

# Build the Docker image
docker build -t document-reader-mcp:latest .

See Docker Configuration below for MCP client setup.

Running the Server

After installation, start the MCP server:

python -m server.main

The server runs over stdio for integration with MCP-compatible clients.

Configuration in Cursor (or other MCP clients)

For Cursor IDE

Add this configuration to your Cursor MCP settings:

• macOS/Linux: ~/.cursor/mcp.json

• Windows: %APPDATA%\Cursor\User\globalStorage\mcp.json or via Settings → MCP

macOS/Linux Configuration

{
  "mcpServers": {
    "document-reader": {
      "command": "python3",
      "args": ["-m", "server.main"],
      "cwd": "/absolute/path/to/document-reader-mcp"
    }
  }
}

Windows Configuration

{
  "mcpServers": {
    "document-reader": {
      "command": "python",
      "args": ["-m", "server.main"],
      "cwd": "C:\\Users\\YourUsername\\document-reader-mcp"
    }
  }
}

Important for Windows users:

• Use double backslashes (\\) in JSON paths, or use forward slashes (/) which also work on Windows

• Replace YourUsername with your actual Windows username

• Ensure the python command points to your Python 3.10+ installation (check with python --version)

For Claude Desktop or other MCP clients

Add similar configuration to your client's MCP settings file, adjusting the path accordingly.

Docker Configuration

To use the Docker version with MCP clients:

{
  "mcpServers": {
    "document-reader": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "-v", "/absolute/path/to/documents:/documents:ro",
        "document-reader-mcp:latest"
      ]
    }
  }
}

Important notes:

• Replace /absolute/path/to/documents with the directory containing files you want to process

• The -v flag mounts your documents directory as /documents in the container (read-only)

• Use -i for interactive mode (required for stdio communication)

• Use --rm to automatically remove the container after it stops

• File paths in MCP tool calls should use /documents/filename.pdf format

Multiple volume mounts:

If you need to access files from multiple directories:

{
  "mcpServers": {
    "document-reader": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "-v", "/Users/you/Documents:/documents:ro",
        "-v", "/Users/you/Downloads:/downloads:ro",
        "document-reader-mcp:latest"
      ]
    }
  }
}

Custom rate limiting:

{
  "mcpServers": {
    "document-reader": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "-e", "DOC_READER_RATE_LIMIT_PER_MINUTE=120",
        "-v", "/absolute/path/to/documents:/documents:ro",
        "document-reader-mcp:latest"
      ]
    }
  }
}

Security considerations for Docker:

• The container runs as non-root user (UID 1000)

• Volumes are mounted read-only (:ro) for safety

• No network ports are exposed

• Container has minimal attack surface

Available Tools

Once configured, you can use these tools:

Tool: extract_text_from_file

Extract complete text from a document file.

Parameters:

• path (string, required): Absolute or relative path to the document

• max_pages (int, optional): For PDFs, parse only the first N pages (default: 50, set to 0 to disable)

• max_rows (int, optional): For CSV/Excel, parse only N data rows (default: 500, set to 0 to disable)

Returns: Extracted text as string (automatically truncated at 100,000 characters by default)

Supported formats: .pdf, .xlsx, .xlsm, .csv, .txt, .json, .md, .docx

Note: For large files, use extract_text_from_file_stream instead to avoid memory issues.

Default Limits: To prevent AI context overflow, the tool applies sensible defaults:

• PDFs: First 50 pages

• Excel/CSV: First 500 rows

• All formats: 100,000 character output limit

Tool: extract_text_from_file_stream

Stream text chunks from a document (memory-efficient for large files).

Parameters:

• path (string, required): Absolute or relative path to the document

• max_pages (int, optional): For PDFs, page cap (default: 50, set to 0 to disable)

• max_rows (int, optional): For CSV/Excel, row cap (default: 500, set to 0 to disable)

• chunk_size (int, optional): Characters per chunk (default: 4096, min: 512)

Yields: Text chunks as strings

Supported formats: All formats from extract_text_from_file

Tool: convert_to_markdown

Convert various document formats to Markdown, extracting and saving images when applicable.

⚠️ Important: This tool converts the ENTIRE document and saves it to a file. It ignores the DOC_READER_DEFAULT_MAX_ROWS, DOC_READER_DEFAULT_MAX_PAGES, and DOC_READER_MAX_OUTPUT_CHARS environment variables. Only the preview returned to the AI is limited to protect context - the saved file contains the complete document.

Parameters:

• path (string, required): Absolute or relative path to the file to convert

• output_dir (string, optional): Directory where the markdown file and images will be saved. If not specified, saves in the same directory as the source file

• output_filename (string, optional): Name for the output markdown file (without extension). If not specified, uses the source filename with .md extension

Returns: Dictionary containing:

• markdown_path: Path to the saved markdown file (contains FULL content, not truncated)

• images_dir: Path to the directory containing extracted images (if any)

• image_count: Number of images extracted

• markdown_preview: First 500 characters preview (truncated for AI context protection)

• file_size_chars: Total character count of the saved markdown file

• status: "success" or error status

• message: Human-readable status message

Supported formats:

• PDF (.pdf) - with automatic image extraction and positioning at page locations

• Excel (.xlsx, .xlsm, .xltx, .xltm) - converted to markdown tables

• Word (.docx) - with image extraction

• CSV (.csv) - converted to markdown tables

• PowerPoint (.pptx) - text and images

• HTML (.html, .htm)

• Plain text (.txt, .log)

• Images (.jpg, .jpeg, .png) - with OCR if available

Example Usage:

# Convert a Word document with images
result = convert_to_markdown(
    path="/path/to/document.docx",
    output_dir="/path/to/output"
)
# Creates: /path/to/output/document.md
#          /path/to/output/document_images/image_1.png
#          /path/to/output/document_images/image_2.png

Important Notes:

• Full file is saved: The complete markdown file is saved to disk without any truncation, regardless of size

• Preview is truncated: Only the preview returned to the AI is limited to 500 characters to protect context

• Images: Automatically extracted from supported formats and saved in a {filename}_images/ subdirectory, with markdown using relative paths to reference them

• PDF images: Images are intelligently positioned throughout the markdown document at their corresponding page locations, making them viewable in preview

Usage Examples

In Cursor Chat:

Extract text from ~/Downloads/report.pdf and summarize the findings

Read the CSV file data.csv and show me the first 10 rows

What's in the JSON file config.json?

Convert the Word document ~/Documents/proposal.docx to Markdown and save it in ~/Documents/markdown/

Convert this Excel file to Markdown: ~/data/sales_report.xlsx

Programmatic Usage:

# Via MCP client - Extract text
result = await client.call_tool("extract_text_from_file", {
    "path": "/path/to/document.pdf",
    "max_pages": 5
})

# Streaming large files
async for chunk in client.stream_tool("extract_text_from_file_stream", {
    "path": "/path/to/large_file.csv",
    "chunk_size": 8192
}):
    print(chunk)

# Convert to Markdown
result = await client.call_tool("convert_to_markdown", {
    "path": "/path/to/document.docx",
    "output_dir": "/path/to/output",
    "output_filename": "converted_document"
})
print(f"Markdown saved to: {result['markdown_path']}")
print(f"Images extracted: {result['image_count']}")

Configuration

Environment Variables

Configure the server behavior using these environment variables:

• DOC_READER_RATE_LIMIT_PER_MINUTE: Maximum tool calls per minute (default: 60)

  • Applies to: All tools

• DOC_READER_MAX_OUTPUT_CHARS: Maximum output text size in characters (default: 100000)

  • Applies to: extract_text_from_file and extract_text_from_file_stream only

  • Does NOT apply to: convert_to_markdown (saves full file, only preview is limited)

• DOC_READER_DEFAULT_MAX_ROWS: Default maximum rows for spreadsheets/CSV (default: 500, set to 0 to disable)

  • Applies to: extract_text_from_file and extract_text_from_file_stream only

  • Does NOT apply to: convert_to_markdown (converts entire document)

• DOC_READER_DEFAULT_MAX_PAGES: Default maximum pages for PDFs (default: 50, set to 0 to disable)

  • Applies to: extract_text_from_file and extract_text_from_file_stream only

  • Does NOT apply to: convert_to_markdown (converts entire document)

Example:

export DOC_READER_RATE_LIMIT_PER_MINUTE=120
export DOC_READER_MAX_OUTPUT_CHARS=200000
export DOC_READER_DEFAULT_MAX_ROWS=1000
export DOC_READER_DEFAULT_MAX_PAGES=100
python -m server.main

Why these limits? Large documents can easily exceed AI model context windows (typically 200K-1M tokens). These defaults prevent context overflow while allowing flexibility for specific use cases. When limits are hit, the tool provides clear warnings with instructions on how to adjust them.

Technical Details

File Size Limits

• Maximum file size: 100 MB

• Files larger than this will be rejected with an error

Encoding Detection

Text-based formats (CSV, TXT, JSON, Markdown) automatically try multiple encodings:

• UTF-8

• Latin-1 (ISO-8859-1)

• Windows-1252 (CP1252)

Dependencies by Format

Security Considerations

⚠️ Important: This server reads local files from the filesystem.

• Do NOT expose this server to untrusted networks

• Only use in trusted MCP client environments (e.g., Cursor IDE)

• Rate limiting is per-process, not per-user

• No authentication is built-in

• File paths are expanded with os.path.expanduser() (supports ~)

Troubleshooting

"Unsupported file type" error

• Check that the file extension matches one of the supported formats

• Supported: .pdf, .xlsx, .xlsm, .xltx, .xltm, .docx, .csv, .txt, .log, .json, .md, .markdown

"Failed to decode" error

• The file may use an unsupported text encoding

• Try converting the file to UTF-8 encoding first

• This typically affects CSV, TXT, JSON, and Markdown files

Rate limit exceeded

• Increase the DOC_READER_RATE_LIMIT_PER_MINUTE environment variable

• Or wait 60 seconds for the rate limit window to reset

Missing dependency errors

• If you see "X is not installed" errors, reinstall dependencies: pip install -r requirements.txt

• For PDF image extraction issues, ensure PyMuPDF is installed: pip install pymupdf

Windows-specific issues

PowerShell execution policy error

If you see cannot be loaded because running scripts is disabled:

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

Path length limitations (Windows)

Windows has a 260-character path limit by default. For long paths:

1. Enable long path support in Windows 10/11: Microsoft Docs

2. Or move the repository to a shorter path (e.g., C:\mcp\document-reader)

Python not found on Windows

• Ensure Python 3.10+ is installed and added to PATH

• Verify with: python --version

• If python doesn't work, try py or python3

Virtual environment activation issues on Windows

• Command Prompt: Use .venv\Scripts\activate.bat

• PowerShell: Use .venv\Scripts\Activate.ps1

• Git Bash: Use source .venv/Scripts/activate

Docker-related issues

Docker not running

• Ensure Docker Desktop is installed and running

• On Windows, Docker Desktop requires WSL 2

Permission errors with Docker volumes

• On Windows, ensure the drive is shared in Docker Desktop settings

• Right-click Docker Desktop icon → Settings → Resources → File Sharing

Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines on:

• Setting up your development environment

• Code style and commit conventions

• Adding support for new file formats

• Submitting pull requests

License

This project is licensed under the MIT License - see the LICENSE file for details.

Support

• Issues: GitHub Issues

• Discussions: GitHub Discussions

Version

Current version: 1.0.0