docx-mcp

What is docx-mcp?

docx-mcp is an MCP server for reading and editing Word (.docx) documents with full support for track changes, comments, footnotes, and OOXML-level structural validation. It is designed for developers and AI agents who need programmatic control over .docx files.

How to use docx-mcp?

Install via pip (pip install docx-mcp-server) or uvx (uvx docx-mcp-server). Configure the server in your MCP client’s settings JSON (e.g., Claude Desktop, Cursor, VS Code), then use the provided tools such as open_document, insert_text, and save_document to interact with .docx files.

Key features of docx-mcp

• Track changes with proper w:ins/w:del markup for revisions
• Add, reply to, and read threaded comments
• Add, list, and validate footnotes
• Validate paraId uniqueness across all document parts
• Detect and remove VML watermarks (e.g., DRAFT)
• Perform comprehensive structural audits (footnotes, headings, bookmarks, images)

Use cases of docx-mcp

• Editing contracts with tracked changes and client comments
• Automating document revision workflows with revision marks
• Auditing document structure for heading and footnote consistency
• Searching and replacing text across body, footnotes, and comments
• Removing draft watermarks from document headers programmatically

FAQ from docx-mcp

What are the system requirements?

Python 3.10+ and the lxml library are required.

How does docx-mcp differ from python-docx?

docx-mcp manipulates OOXML directly, giving full control over track changes (w:ins/w:del), comments, and structural validation that python-docx does not expose.

Can I save changes to a new file?

Yes, the save_document tool supports saving to a new path or overwriting the original file.

What data does the server access?

Only the .docx files you explicitly open via the open_document tool; no remote data is accessed.

Is there authentication or a specific transport?

No authentication is required; the server runs locally over standard I/O as an MCP server.