MCP Docling

🚀 MCP Document Processing Server

An MCP server that uses the Docling library to provide document processing capabilities.

🚀 Quick Start

📦 Installation

You can install the server using pip:

pip install -e .

💻 Usage Examples

Basic Usage

Start the server using standard input/output (default) or SSE transmission:

# Use stdio transmission (default)
mcp-server-lls

# Use SSE transmission and a custom port
mcp-server-lls --transport sse --port 8000

If you use uv, you can run the server directly without installation:

# Use stdio transmission (default)
uv run mcp-server-lls

# Use SSE transmission and a custom port
uv run mcp-server-lls --transport sse --port 8000

Available Tools

The server exposes the following tools:

1. convert_document: Convert a document from a URL or local path to Markdown format.

   • source: URL or local file path of the document (required).
   • enable_ocr: Whether to enable OCR processing for scanned documents (optional, default: false).
   • ocr_language: List of OCR language codes, e.g., ["en", "fr"] (optional).

2. convert_document_with_images: Convert a document and extract embedded images.

   • source: URL or local file path of the document (required).
   • enable_ocr: Whether to enable OCR processing for scanned documents (optional, default: false).
   • ocr_language: List of OCR language codes (optional).

3. extract_tables: Extract tables from a document and return them as structured data.

   • source: URL or local file path of the document (required).

4. convert_batch: Process multiple documents in batch.

   • sources: List of document URLs or file paths (required).
   • enable_ocr: Whether to enable OCR processing for scanned documents (optional, default: false).
   • ocr_language: List of OCR language codes (optional).

5. qna_from_document: Convert a document to a YAML-formatted Q&A file.

   • source: URL or local file path of the document (required).
   • no_of_qnas: Expected number of Q&A pairs (optional, default: 5).

   ⚠️ Important Note

   This tool requires IBM Watson X credentials. Set the following environment variables:

   • WATSONX_PROJECT_ID: Your Watson X project ID.
   • WATSONX_APIKEY: Your IBM Cloud API Key.
   • WATSONX_URL: Watson X API address (default: https://us-south.ml.cloud.ibm.com).

6. get_system_info: Get information about system configuration and acceleration status.

Example of Using with Llama Stack

You can integrate this server with Llama Stack to provide document processing capabilities for your large language model application. Make sure you have a running Llama Stack instance.

# Here is an example code:
_summary_and_qna('https://arxiv.org/pdf/2004.07606')

Cache

The server caches processed documents in the ~/.cache/mcp-docling/ directory to improve the performance of repeated requests.