Docling MCP

Docling is an open-source document reader started by IBM Research and now run under the Linux Foundation's AI & Data group. It's very good at the hard parts of PDFs: tables, multi-column layouts and scanned pages. Its MCP server lets Claude use it, so "pull the pricing table out of this supplier PDF" comes back as a proper table.

It needs uv. The command below runs everything on your own computer, which keeps documents private. The first run downloads Docling's models, so it's slow to start and heavy on older laptops. Teams with a Docling Serve server can point it there instead.

It's listed in Claude's connector directory for the desktop app too. For everyday PDFs Claude can often read them itself. Docling earns its place on long, messy or scanned documents.

• Convert a PDF or document from your computer or a link into structured text
• Keep tables, headings and layout instead of a jumble of words
• Read scanned pages with OCR (text recognition)
• Export the result as Markdown
• Write new documents section by section and save them

Open your terminal (the text window where you run Claude Code) in your project folder, then paste this command to connect it:

claude mcp add docling -e DOCLING_MCP_CONVERSION_MODE=local -- uvx --from "docling-mcp[local]" docling-mcp-server

Or add this to your project's .mcp.json:

{
  "mcpServers": {
    "docling": {
      "command": "uvx",
      "args": ["--from", "docling-mcp[local]", "docling-mcp-server"],
      "env": { "DOCLING_MCP_CONVERSION_MODE": "local" }
    }
  }
}

People who deal with long, messy PDFs: contracts, reports, supplier price lists and scanned paperwork.

• Pulling tables out of PDFs
• Scanned documents
• Long reports
• Preparing documents for search