ChatGPT can read Word documents and discuss their content, but it cannot natively produce a new .docx file with tracked changes. When you upload a contract or agreement to ChatGPT and ask it to redline something, the output is text in the chat window — not a downloadable Word file with insertions and deletions marked up in Word's Review format. Closing that gap requires either Microsoft's Copilot for M365 (if you are inside the Microsoft ecosystem) or a tool like the Scaffold MCP connector (if you want Word-editing capabilities inside ChatGPT without an enterprise Microsoft subscription).
This article explains what ChatGPT can and cannot do with Word documents today, how those options compare to Copilot for M365, and how Scaffold fills the remaining gap.
What Can ChatGPT Do with a Word Document Natively?
ChatGPT supports .docx file uploads on ChatGPT Plus and Team plans. When you upload a Word document, ChatGPT reads the document's text content and makes it available for the conversation. From there, ChatGPT can do quite a lot:
Reading and analysis. ChatGPT can summarize a document, identify specific clauses or sections, answer questions about the content, and flag language that appears one-sided, ambiguous, or inconsistent with stated requirements. For a 30-page vendor agreement, this kind of read-through takes ChatGPT seconds rather than the hour a human reviewer might spend on a first pass.
Suggesting changes. You can ask ChatGPT to rewrite a clause, propose alternative language for a provision, or add a missing obligation — and ChatGPT will produce revised text in the chat window. This is useful as a drafting starting point.
Explaining provisions. For non-lawyers dealing with contracts, ChatGPT can explain what a clause means in plain English, which provisions are standard vs. unusual, and what the practical implications of specific language are.
What ChatGPT cannot do natively is write those changes back into the Word document as tracked changes. Every suggestion exists only in the chat window. You have to open Word, find the relevant section, and type or paste the change yourself. For a document with ten suggested changes across thirty pages, that manual work adds up and introduces the risk of missing a change or misapplying it.
What About ChatGPT Plugins or Actions?
OpenAI has experimented with various plugin and action ecosystems since 2023. As of 2026, the relevant integration path for Word document editing in ChatGPT is the MCP (Model Context Protocol) connector system, which ChatGPT Plus supports. MCP connectors — including the Scaffold MCP connector — are the mechanism that gives ChatGPT external tool access during a conversation, including the ability to read documents, apply changes, and return downloadable files.
There is no native OpenAI plugin for producing Word tracked changes. The Scaffold MCP connector is the practical way to add that capability to ChatGPT.
How Copilot for M365 Compares
Microsoft Copilot for M365 is worth understanding as the baseline enterprise option. If your organization pays for Microsoft 365 and has added the Copilot license, you have an AI built directly into Word desktop that can read your document, suggest changes, and apply some edits inline.
Copilot for M365 vs. Scaffold MCP: The Key Differences
Copilot for M365 runs inside the Word desktop application and has deep integration with the document object model — it can read styles, formatting, tables, and comments with full fidelity, and it can write changes back to the document in ways that Word natively understands. The tradeoff is the requirement stack: Copilot requires Word desktop (not Word Online for full functionality), a Microsoft 365 subscription, and the Copilot add-on license at approximately $30 per user per month on top of M365. For organizations that already run M365, this is a reasonable investment. For individuals or small teams without an M365 subscription, or for anyone who primarily works in ChatGPT rather than in Word, it is not a practical path. The Scaffold MCP connector approaches the same problem from the other direction: instead of adding AI to Word, it adds Word capabilities to the AI you already use. It works in ChatGPT Plus, Claude, and other MCP-compatible AI clients, requires no desktop software, and costs $29/month — making it accessible to individuals and small teams who want professional-grade output without enterprise licensing overhead.
How the Scaffold MCP Connector Works in ChatGPT
The Scaffold MCP connector is an MCP server built for Word document operations. You connect it to ChatGPT through the ChatGPT settings once, and it becomes available in every subsequent conversation. When you upload a Word document and ask ChatGPT to make specific changes, ChatGPT calls Scaffold's tools to read the document, determine the changes, apply them as proper tracked-change XML, and return a download link.
The output is a .docx file with every proposed change embedded as a tracked revision — deletions in strikethrough, insertions underlined, each attributed to the reviewer — exactly the format you would send to a client or counterparty for review. The recipient opens it in Word and uses the Review pane to accept or reject each change individually.
A typical conversation might look like this: you upload a draft NDA, type "redline this to make the confidentiality obligation mutual and extend the term from one year to two years," and ChatGPT — with Scaffold connected — returns a download link to a revised .docx with two tracked changes applied. You download, review, and send.
Frequently Asked Questions
Can I use Scaffold with ChatGPT Free?
The Scaffold MCP connector works through ChatGPT's MCP integration, which as of 2026 requires a ChatGPT Plus or Team subscription. The ChatGPT Free tier does not support MCP connectors. If you are on ChatGPT Free, you can still use Scaffold through Claude (claude.ai Pro), which also supports MCP connectors. Scaffold's 7-day free trial works with either AI client.
What is the difference between ChatGPT's native .docx upload and Scaffold MCP?
ChatGPT's native file upload lets ChatGPT read a Word document and return suggestions as text in the chat window. Scaffold MCP lets ChatGPT read the document, apply the suggested changes as tracked revisions inside the Word file, and return a downloadable .docx. The output type is the key difference: text vs. a properly formatted Word document with an embedded change history.
Does Scaffold work if I do not have Word installed?
Yes. Scaffold processes documents server-side and returns a .docx file. You can download that file and open it in Word (desktop or online), Google Docs, or LibreOffice. You do not need Word installed to produce the redline — only to review or accept the changes afterward.
Can Scaffold handle templates, not just redlining existing documents?
Yes. The Scaffold MCP connector also supports document templates: you define variable fields in a Word template and ask ChatGPT (or Claude) to populate them based on your inputs. This is useful for drafting NDAs, engagement letters, offer letters, or any document with a standard structure and variable facts.
How does Scaffold compare to Spellbook?
Spellbook is a legal AI add-in designed for law firms, priced at $100+/month, and requires installation in Word desktop. It has deep legal-specific training and review features oriented toward law firm workflows. Scaffold is a more general-purpose document tool, starts at $29/month, and works inside ChatGPT or Claude without any desktop install. If you are a solo practitioner or in-house counsel who does not want a legal-specific enterprise subscription, Scaffold is the more practical starting point.
Start a free 7-day Scaffold trial. Setup in ChatGPT takes a few minutes, and you can have a tracked-changes workflow running before the end of your first conversation.