Unstructured MCP Server

Pure JavaScript MCP server for document processing with Unstructured.io API. Process PDFs, Word documents, HTML, images, and more directly from Claude Desktop and other AI clients - no Python required!

Features

• 📄 Multi-format Support: PDF, DOCX, HTML, images (with OCR), and more
• 🚀 NPX Executable: No local installation required
• 🤖 Claude Desktop Integration: Works seamlessly with Claude
• ⚡ Pure JavaScript: No Python dependencies
• 🔍 OCR Support: Extract text from scanned documents
• 📊 Table Extraction: Extract and convert tables
• 📝 Multiple Output Formats: JSON, text, markdown

Quick Start

1. Get API Key

Sign up at https://unstructuredapp.io to get your API key.

2. Run with NPX (No Installation)

# Run directly with NPX
UNSTRUCTURED_API_KEY=your_key_here npx uns-mcp-server

3. Add to Claude Desktop

Add to your Claude Desktop configuration (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "uns-mcp": {
      "command": "npx",
      "args": ["uns-mcp-server"],
      "env": {
        "UNSTRUCTURED_API_KEY": "your_key_here"
      }
    }
  }
}

Installation Options

Global Install

npm install -g uns-mcp-server
uns-mcp-server

Local Project

npm install uns-mcp-server
npx uns-mcp-server

Available Tools

The MCP server provides the following tools:

Document Processing

• process_document - Process any document with OCR and formatting preservation
• extract_text - Extract plain text from documents
• extract_tables - Extract tables in JSON, CSV, or HTML format

Connectors

• list_sources - List configured document sources
• create_source_connector - Create input source (S3, Azure, local, etc.)

Usage Examples

Process a PDF

// In Claude Desktop, use the tool directly
await process_document({
  file_path: "/path/to/document.pdf",
  strategy: "hi_res",
  ocr_enabled: true,
  output_format: "json"
});

Extract Text

await extract_text({
  file_path: "/path/to/document.pdf",
  include_metadata: true
});

Extract Tables

await extract_tables({
  file_path: "/path/to/spreadsheet.xlsx",
  format: "csv"
});

Supported Formats

• Documents: PDF, DOCX, DOC, ODT, RTF, TXT
• Images: PNG, JPG, JPEG, TIFF, BMP (with OCR)
• Web: HTML, XML
• Spreadsheets: XLSX, XLS, CSV
• Presentations: PPTX, PPT
• Email: EML, MSG

Processing Strategies

• auto - Automatically select the best strategy
• hi_res - High resolution processing with layout preservation
• ocr_only - Focus on OCR for scanned documents
• fast - Quick processing for simple documents

Environment Variables

• UNSTRUCTURED_API_KEY - Your Unstructured.io API key (required)
• UNSTRUCTURED_API_URL - Custom API endpoint (optional)
• LOG_LEVEL - Logging level: ERROR, WARN, INFO, DEBUG (default: ERROR)

Testing

Test your installation:

# Create a test file
echo "Hello World" > test.txt

# Process it
npx uns-mcp-server test.txt

Troubleshooting

API Key Issues

# Verify your API key is set
echo $UNSTRUCTURED_API_KEY

# Set it for current session
export UNSTRUCTURED_API_KEY=your_key_here

Connection Issues

• Ensure you have internet connectivity
• Verify the API key is valid
• Check if you're behind a corporate firewall

Changelog

v2.0.2

• Updated GitHub repository to CG-Labs organization
• Documentation improvements

v2.0.1

• Fixed API endpoint to use correct domain
• Improved error handling
• Better connection stability

v2.0.0

• Complete rewrite in pure JavaScript
• Removed all Python dependencies
• Direct API integration
• Improved performance

v1.0.x

• Initial release with Python bridge

License

MIT License

Support

• Issues: GitHub Issues
• GitHub: github.com/CG-Labs/Unstructured-Document-Processor-MCP
• NPM: npmjs.com/package/uns-mcp-server
• Unstructured.io: Documentation

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Made with ❤️ for the Claude Desktop community