docling-project/docling-mcp

What it solves

Docling MCP allows AI agents and LLM clients to programmatically convert, process, and generate structured documents. It bridges the gap between raw PDF files and AI-ready structured data by providing a standardized set of tools that an agent can call to handle document conversion and RAG workflows.

How it works

The project implements the Model Context Protocol (MCP), enabling it to act as a server that exposes document processing tools to compatible AI clients (like Claude for Desktop or LM Studio). It leverages the Docling library to transform PDFs into a structured JSON format called DoclingDocument.

Users can run the service in three modes:

• Remote Mode: Connects to a Docling Serve API for lightweight operation.
• Local Mode: Performs all conversions locally on the user's machine.
• Hybrid Mode: Uses a remote service with an automatic fallback to local processing.

Who it’s for

• AI Agent Developers: Those building agents that need to read and structure complex PDFs.
• LLM Users: People using MCP-compatible desktop clients who want their AI to be able to convert and generate documents.
• RAG Developers: Developers integrating document processing with vector databases like Milvus or frameworks like LlamaIndex and LlamaStack.

Highlights

• Agentic Document Control: Tools for creating, editing (adding titles, headings, paragraphs, and nested lists), and exporting documents to Markdown.
• Flexible Deployment: Supports multiple transport protocols including stdio, sse, and streamable-http.
• RAG Integration: Built-in support for Milvus upload/retrieval and integration with LlamaIndex and LlamaStack.
• Performance Features: Includes local document caching and memory management for large files.
• OCR Support: Integrated OCR and table structure detection for high-fidelity conversion.