docx-mcp

Give your AI coding agent the ability to create, read, and edit Word documents. Edits appear as tracked changes in Microsoft Word — red strikethrough for deletions, green underline for insertions — and after any revision session your agent can produce an email-ready change log listing every insertion, deletion, and replacement. Or compare two separate files to get the same output automatically.

Who This Is For

Professionals who produce Word deliverables and want their AI agent to handle the document work directly:

• Legal — contract review with tracked redlines, batch clause replacement across templates, comment annotations explaining each change, footnote management
• Security & Penetration Testing — generate pentest reports from markdown findings, merge appendices from multiple engagements, add executive-summary comments, remove DRAFT watermarks before delivery
• Consulting — build proposals and SOWs from templates, convert meeting notes to formatted deliverables, bulk-update payment terms across document sets
• Compliance & Audit — structural validation of document integrity, cross-reference checking, heading-level audits, protection enforcement

What You Can Ask Your Agent To Do

Review a contract:

"Open contract.docx, find every instance of 'Net 30', change it to 'Net 60', and add a comment on each explaining it was updated per Amendment 3. Save as contract_revised.docx."

Generate a report from markdown:

"Convert my pentest-findings.md to a Word document using the client's report template. Add footnotes for each CVE reference."

Batch-edit a template library:

"Open the MSA template, replace 'ACME Corp' with 'GlobalTech Inc' everywhere, update the effective date in the header, and set document protection to track-changes-only."

Audit a document before sending:

"Open the final deliverable, run a structural audit, check for any DRAFT or TODO markers, validate all footnote cross-references, and remove any watermarks."

Your agent handles the entire workflow — opening the file, navigating the structure, making precise edits with full revision history, validating integrity, and saving — while you focus on the substance.

Installation

Claude Code (recommended):

claude mcp add docx-mcp -- uvx docx-mcp-server

That's it. A companion skill auto-installs the first time the server starts, teaching Claude the document editing workflow, OOXML pitfalls, and audit checklist. It auto-updates on every upgrade.

{
  "mcpServers": {
    "docx-mcp": {
      "command": "uvx",
      "args": ["docx-mcp-server"]
    }
  }
}

{
  "mcpServers": {
    "docx-mcp": {
      "command": "uvx",
      "args": ["docx-mcp-server"]
    }
  }
}

mcpServers:
  docx-mcp:
    command: uvx
    args:
      - docx-mcp-server

pip install docx-mcp-server

Capabilities

Example: Contract Review with Redlines

1. open_document("services-agreement.docx")
2. get_headings()                              → see document structure
3. search_text("30 days")                      → find the payment clause
4. delete_text(para_id, "30 days")             → tracked deletion  (red strikethrough)
5. insert_text(para_id, "60 days")             → tracked insertion (green underline)
6. add_comment(para_id, "Extended per client request — see Amendment 3")
7. audit_document()                            → verify structural integrity
8. save_document("services-agreement_redlined.docx")

Open the output in Word and you see exactly what a human reviewer would produce — revision marks, comments in the margin, clean document structure.

Example: Change Log After a Revision Session

1. open_document("contract.docx")
2. replace_text(para_id, "30 days", "60 days")           → tracked revision (del + ins)
3. add_comment(para_id, "Extended per Amendment 3")
4. save_document("contract_redlined.docx")
5. generate_change_summary("contract_changes.txt")

The .txt output reads like:

1. REPLACEMENT by Claude on 2026-05-25
   Deleted:  "30 days"
   Inserted: "60 days"

Paste it into an email, a PR description, or a handover note.

Example: Compare Two Separate Document Versions

No editing session required — point it at any two files:

1. diff_to_text("proposal_v1.docx", "proposal_v2.docx")

Returns two files:

• proposal_v1_diff.docx — open in Word to see tracked changes inline
• proposal_v1_diff.txt — plain-text summary, ready to paste anywhere

Example: Pentest Report from Markdown

1. create_from_markdown("pentest-report.docx",
       md_path="findings.md",
       template_path="client-template.dotx")
2. audit_document()                            → verify integrity
3. save_document()                             → ready for delivery

Your markdown findings — headings, tables of affected hosts, code blocks with proof-of-concept output, severity ratings — become a formatted Word document matching the client's template. Smart typography is applied automatically (curly quotes, em dashes, proper ellipses).

PII Scrubbing (Experimental)

This feature is experimental and NOT suitable for production use or as the sole control for privileged, regulated, or legally sensitive documents.

scrub_pii uses Microsoft Presidio with a spaCy NER model (en_core_web_lg, ~560 MB, downloaded on first call) to detect and permanently redact personal information. Redacted spans are replaced with solid black DrawingML rectangles — the original text is deleted from the XML, not merely hidden.

What it detects reliably (via regex/pattern matching, high confidence):

• Email addresses, phone numbers, credit card numbers, SSNs, IBANs, IP addresses, US passport numbers

What it detects statistically (via NER, will have misses):

• Person names, organisation names, locations, dates

Known NER gaps — these will be missed:

• Names in ALL-CAPS (common in ledgers, table headers, legal party designations)
• Single-token names with no surrounding context
• Non-English names (Arabic, CJK, African name patterns have low recall on this English model)
• Names embedded in legal boilerplate (Lender: Jane Doe, Authorized Signatory: John Smith)

Recommended workflow: