Docling MCP

Name: Docling MCP
Author: Docling Project

Turns PDFs, Office files and web pages into structured Docling documents, and builds or edits them

Official · Docling Project Python Local · stdio Productivity AI and ML

GitHub PyPI Website

43 popularity score ★ 787

Connect your client

Local (PyPI)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio Add to Goose

Opens the app, which asks before it adds the server. Then replace each <NAME> in its settings with your value.

Or add it by command or config

It needs DOCLING_MCP_SERVICE_URL. Replace each <NAME> with your value.

Claude CodeTerminal

claude mcp add docling -e DOCLING_MCP_SERVICE_URL=<DOCLING_MCP_SERVICE_URL> -- uvx --from=docling-mcp docling-mcp-server --transport stdio

Claude Desktopclaude_desktop_config.json

{
  "mcpServers": {
    "docling": {
      "command": "uvx",
      "args": [
        "--from=docling-mcp",
        "docling-mcp-server",
        "--transport",
        "stdio"
      ],
      "env": {
        "DOCLING_MCP_SERVICE_URL": "<DOCLING_MCP_SERVICE_URL>"
      }
    }
  }
}

Cursor~/.cursor/mcp.json

{
  "mcpServers": {
    "docling": {
      "command": "uvx",
      "args": [
        "--from=docling-mcp",
        "docling-mcp-server",
        "--transport",
        "stdio"
      ],
      "env": {
        "DOCLING_MCP_SERVICE_URL": "<DOCLING_MCP_SERVICE_URL>"
      }
    }
  }
}

VS Code.vscode/mcp.json

{
  "servers": {
    "docling": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "--from=docling-mcp",
        "docling-mcp-server",
        "--transport",
        "stdio"
      ],
      "env": {
        "DOCLING_MCP_SERVICE_URL": "<DOCLING_MCP_SERVICE_URL>"
      }
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.docling]
command = "uvx"
args = ["--from=docling-mcp", "docling-mcp-server", "--transport", "stdio"]

[mcp_servers.docling.env]
DOCLING_MCP_SERVICE_URL = "<DOCLING_MCP_SERVICE_URL>"

Gemini CLITerminal

gemini mcp add --scope user -e DOCLING_MCP_SERVICE_URL=<DOCLING_MCP_SERVICE_URL> docling uvx -- --from=docling-mcp docling-mcp-server --transport stdio

Settings 10

`DOCLING_MCP_SERVICE_URL`	Required: URL of the Docling Serve instance that does the parsing in remote mode, the default; not needed with DOCLING_MCP_CONVERSION_MODE=local.
`DOCLING_MCP_SERVICE_API_KEY`	Optional: API key for the Docling Serve instance, if it asks for one.
`DOCLING_MCP_CONVERSION_MODE`	Optional: remote (default) or local; local needs the docling-mcp[local] install.
`DOCLING_MCP_FALLBACK_TO_LOCAL`	Optional: Switch to local parsing when the remote service is unreachable; false by default, needs docling-mcp[local].
`DOCLING_MCP_SERVICE_TIMEOUT`	Optional: Seconds allowed for one remote conversion job; 300 by default.
`DOCLING_MCP_DO_OCR`	Optional: Run OCR; true by default.
`DOCLING_MCP_DO_TABLE_STRUCTURE`	Optional: Detect table structure; true by default.
`DOCLING_MCP_DO_PICTURE_DESCRIPTION`	Optional: Describe figures with a vision-language model; false by default and slow on CPU.
`DOCLING_MCP_IMAGE_EXPORT_MODE`	Optional: How images appear in Markdown: placeholder (default), embedded or referenced.
`DOCLING_MCP_CACHE_MAX_DOCUMENTS`	Optional: Documents kept in the in-memory cache; 10 by default.

What it does

Docling MCP puts the Docling document parser behind MCP tools. Hand it a local path or a URL and it parses the file (PDF, DOCX, HTML, images and the other formats Docling reads) into a DoclingDocument, keeps that document in an in-memory cache and returns a key. Later calls use the key to export Markdown, render a page thumbnail, search the text, or rewrite and delete individual items.

The same server can also write documents from scratch: an assistant creates an empty document, then adds a title, section headings, paragraphs, nested lists and HTML tables step by step, and finally saves it as Markdown and JSON.

Parsing happens in one of two places. In remote mode, the default, files go to a Docling Serve instance you point it at. In local mode, Docling runs on your own machine, which needs the larger docling-mcp[local] install. A fallback setting lets remote mode switch to local when the service cannot be reached.

By default it loads three tool groups: conversion, generation and manipulation, 21 tools in all (counted in the v3.3.0 source). Three more groups can be turned on with --tools: llama-index-rag (Milvus indexing and retrieval through LlamaIndex), llama-stack-rag and llama-stack-ie (LlamaStack retrieval and structured extraction). The project sits under the LF AI & Data Foundation and was started at IBM Research Zurich.

What it needs

Runtime	Python

uv (for uvx) and Python. For remote mode, the default: a running Docling Serve instance, set with DOCLING_MCP_SERVICE_URL, plus its API key if it has one. For local mode: install docling-mcp[local] and set DOCLING_MCP_CONVERSION_MODE=local; Docling then runs its models on your machine. The optional RAG groups need their extras (llama-index-rag, llama-stack) and an LLM endpoint or LlamaStack server.

Limits

v3.3.0 starts on streamable-http at localhost:8000 unless --transport stdio is given (servers/mcp_server.py), so the README's client config, which has no --transport, does not speak stdio; the command here adds it. In remote mode with no DOCLING_MCP_SERVICE_URL and no fallback, the conversion tools fail; generation and manipulation still work. The directory tool converts only files directly in the folder, not subfolders, and skips files that fail. The cache keeps 10 documents by default (DOCLING_MCP_CACHE_MAX_DOCUMENTS), dropping the least recently used. Version 3 needs MCP SDK v2 clients; the README says to pin docling-mcp<3.0.0 otherwise.

Who it suits

Suits developers who want an assistant to read real documents, such as scanned PDFs, tables and slides, as structure rather than raw text, and who either have a Docling Serve instance or are willing to run Docling locally with the heavier install. Local mode keeps files on your own machine; remote mode sends them to whichever service you set. It also suits workflows where the assistant drafts a report and saves it as Markdown and JSON. For a quick text dump of a single file, a simpler file reader is lighter than running Docling's models.

Tools 21

convert_document_into_docling_document: Parses a file path or URL into a cached Docling document and returns its key
convert_directory_files_into_docling_document: Parses every file directly inside a local folder into cached documents
check_document_in_local_cache: Checks whether a document key is in the cache
remove_document_from_local_cache: Removes one document from the cache
list_cached_documents: Lists the keys of all cached documents
create_new_docling_document: Starts a new empty document and returns its key
export_docling_document_to_markdown: Returns a cached document as Markdown, optionally cut to a maximum size
save_docling_document: Writes a cached document to Markdown and JSON files in the cache folder
get_page_thumbnail: Renders one page of a document as a PNG image
add_title_to_docling_document: Adds or replaces the document title
add_section_heading_to_docling_document: Adds a section heading at a given level
add_paragraph_to_docling_document: Adds a paragraph of text
open_list_in_docling_document: Opens a new list, nested if one is already open
close_list_in_docling_document: Closes the list that is currently open
add_list_items_to_list_in_docling_document: Adds items to the open list
add_table_in_html_format_to_docling_document: Adds a table given as HTML
get_overview_of_document_anchors: Shows the document's structure with an anchor for each item
search_for_text_in_document_anchors: Finds items whose text matches given keywords
get_text_of_document_item_at_anchor: Returns the text of the item at an anchor
update_text_of_document_item_at_anchor: Replaces the text of the item at an anchor
delete_document_items_at_anchors: Deletes the items at the given anchors

Popularity

43 popularity score, on our list (the bar is 36)

★ 787

GitHub stars	787	34.8 points
PyPI downloads, last month	0
Official server	Yes	8 points

Measured 2026-11-07. How the score works.

Where it is published

Code	docling-project/docling-mcp ★ 787
PyPI	docling-mcp-server
PyPI	docling-mcp

What it can reach

Runs	On your machine
Reaches	Files on your machine, The internet
Changes things	Yes: it can write, send, create or delete
Sign-in	An API key or token
Licence	MIT (GitHub)
Last code change	2026-11-04

Runs on your machine as a Python process. The conversion tools read any local file or folder path, or download any URL, given to them, with the permissions of the user running the server. In remote mode each document is sent to the Docling Serve URL you set, using DOCLING_MCP_SERVICE_API_KEY if one is given; in local mode parsing stays on your machine. It writes Markdown, JSON and PNG files into its cache folder (save_docling_document, get_page_thumbnail), and the manipulation tools change or delete items in cached documents. With --transport sse or streamable-http it listens on localhost:8000 unless --host or --port is given. The optional RAG groups send text to the LLM, embedding and vector endpoints you configure.

Over time

GitHub stars 787 +15 (+2 %)

787772

14 Oct 20267 Nov 2026

PyPI downloads a month 0 +0

14 Oct 20267 Nov 2026

Popularity score 43 +0

4343

14 Oct 20267 Nov 2026

What is rising across the directory

Sources

Page checked 2026-10-14.

Badge for your README

Shows this server's rank in its category, and stars for a server that runs on your machine. It updates every hour.

MarkdownREADME.md

[![Docling MCP on EveryMCPServer](https://everymcpserver.com/badge/docling-mcp.svg)](https://everymcpserver.com/servers/docling-mcp)

Connect your client

Add in one click

Or add it by command or config

Settings 10

What it does

What it needs

Limits

Who it suits

Tools 21

Similar servers

AFFiNE MCP Server

Discourse MCP

kordoc

MCP WordPress

Box MCP server

OpenWork MCP Gateway