Use an MCP to edit Word docs with tracked changes
An agent can revise .docx files by editing HTML first, then translating the result back into Word’s file format.
- What they did
- A custom-built agent was wired to an MCP for Word editing. It worked on an HTML version of the document rather than raw OOXML, and a small 3–8B LoRA model converted each altered block into OOXML. A deterministic matcher found the right block, then the system created tracked edits by comparing versions and patching the .docx.
- How it went
- On the vendor’s 279-task benchmark, the approach came out 2.7–2.9x less expensive and 2.7–3.5x quicker than Anthropic’s DOCX skill, with a better pass rate. Hyperlink handling lagged, at 25% pass rate. Comments, image attachments, and latent styles are not supported yet.
- Tips
- This fits documents where layout needs to stay intact and changes should be reviewable. Be cautious about the benchmark, since it is the vendor’s own.
Try it yourself with Claude Cowork
Load [file path] in Word and carry out these edits: [describe the changes]. Preserve the existing formatting, and show the edits as tracked changes. Save a separate copy named [new filename], not the original. Before any step that would permanently change, remove, send, or submit the file, stop and ask for my approval. When you’re done, summarize each edit you made.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Turn Gmail threads into a clean to-do list
A Chrome extension read Gmail and pulled action items from emails, invites, and follow-ups into a prioritized list.
Summarize today’s email and line up meetings
They connected an assistant to mail, calendar, meetings, and other apps, then asked it to turn the day’s inbox into a fast, usable brief.
Audit your mail DNS and add DMARC
An agent reviewed a domain’s email DNS, spotted a missing DMARC record, and a single TXT update raised the score from 80 to 96.
Use a voice agent for oral exams
A professor used a voice agent to run oral finals, then had an LLM panel grade the transcripts and flag shared weak spots.
Create invoice PDFs, review them, then send after approval
An agent built invoice PDFs for billing or reimbursements, sent the draft to the user, and mailed the final version only after approval.
Turn receipt photos into ledger entries
They used a chat bot to read receipt images, pull out key expense details, and stage them for review in a ledger.