AI-agent editing with WebMCP
WebDecks, WebDocs and WebSheets speak WebMCP, the Web Model Context Protocol, so a browser AI agent can drive your real documents the same way you do. Point ChatGPT's browser mode, or any WebMCP client, at the app: it opens a file, reads the contents, edits them and exports the result. Everything runs in your browser, so your files never leave your device. Building this into your own product? Drive the engine from code with the SDK; WebMCP is the in-browser agent path.
What WebMCP support means
The Web Model Context Protocol lets a web page publish tools that an AI agent calls directly through the browser, via the emerging navigator.modelContext API. WebDecks registers its editing tools with the browser, so a connected agent sees a structured, typed surface and acts on the actual document model. It never guesses from a screenshot.
The agent works through the same engine as the UI, so its edits are real. It reads the true text, shapes and structure of your file, changes them, and exports the finished PPTX, DOCX, XLSX or PDF. There is no lossy round-trip, and nothing uploads to a server. The model runs on your document, on your machine.
Two ways to connect an agent
In-app on the apps themselves. Open WebDecks, WebDocs or WebSheets and the page registers its WebMCP tools with your browser through navigator.modelContext. An agent running in that browser, whether ChatGPT's browser mode or any WebMCP client, can then read and edit whatever document you have open.
The standalone /mcp/ endpoint. A dedicated WebMCP page lives at /mcp/ for agents that want the tool surface directly, without the full editor around it. It exposes the same read, edit and export tools over WebMCP.
The tool surface: read, edit, export
Read. Agents inspect the document model as structured data: slide and page structure, text runs, tables, cell values and formulas, shapes and images. They can answer questions about a file or plan an edit precisely.
Edit. Agents make real changes: insert and rewrite text, fix typos, restyle and format, add or move shapes, update table and spreadsheet cells, and reorganize content. Every change runs through the same engine the editor uses.
Export. When the work is done, an agent can export the result to the native format (PPTX, DOCX, XLSX) or to PDF, exactly as the app's own File menu would.
AI editing, per app
AI editing for presentations — WebDecks
Let an agent open a .pptx, read every slide, fix wording, restyle shapes and re-export to PowerPoint or PDF. A browser-native presentation editor an AI can drive end to end.
Open WebDecks →AI editing for documents — WebDocs
Point an agent at a Word document to read the text model, rewrite and reformat paragraphs, update tables and export back to .docx or PDF.
Open WebDocs →AI editing for spreadsheets — WebSheets
An agent can read cell values and formulas, update ranges, and export an Excel workbook. Spreadsheet automation that runs in the browser, on your own file.
Open WebSheets →On the roadmap: a remote MCP connector
Today WebMCP support runs in the browser, next to your files. A remote HTTP/SSE MCP connector is on the roadmap, so an agent or assistant can reach a WebDecks tool surface over the network as a hosted Model Context Protocol server.
Prefer to drive it from code?
For most products, the SDK is the better path. It is a TypeScript API over the same browser-native engine that powers these apps, so your own agent can load a document, read the true model, edit it and export, all from a few lines of code. WebMCP is for browser agents; the SDK is for your app.
Read the SDK docsThe SDK is free to try. Production use, or shipping on any origin other than your development one, needs a commercial license with an unlocked, origin-bound build.