Reading and appending to a Doc is table stakes; applying a paragraph style, inserting a table at an index, or restyling every occurrence of a phrase is what makes a model useful on a document someone will actually read. The corresponding cost is the setup — a Google Cloud project, a consent screen, and a one-time manual authorization dance that ends in a token file with standing access to your Docs and Drive.
A comprehensive MCP server over the Google Docs and Google Drive APIs, built with `fastmcp`. It goes past reading and appending into real document structure — text styling, paragraph layout, tables and page breaks — and adds full Drive file management on top, with shared-drive support across every operation.
- Document access and editing: `readGoogleDoc` returns plain text, JSON structure or markdown; `appendToGoogleDoc` adds text; `insertText` places it at a specific position; `deleteRange` removes content
- Formatting that goes beyond plain text — `applyTextStyle` for bold, italic and colours, `applyParagraphStyle` for alignment and spacing, and `formatMatchingText` to format by content rather than by index
- Structure: `insertTable` creates tables, `insertPageBreak` inserts breaks, and `fixListFormatting` is offered as an experimental automatic list detector
- Drive discovery — `listGoogleDocs`, `searchGoogleDocs`, `getRecentGoogleDocs` and `getDocumentInfo` for detailed metadata
- Folder and file management: `createFolder`, `listFolderContents`, `getFolderInfo`, `moveFile`, `copyFile`, `renameFile`, `deleteFile`
- Document creation from scratch with `createDocument` or from an existing template with `createFromTemplate`
- Shared-drive support throughout, controlled by a `corpora` parameter — `user` for your personal drive, `drive` for one shared drive with a drive ID, or `allDrives` to search everything you can reach
Your own Google Cloud project with both the Google Docs API and the Google Drive API enabled, an OAuth consent screen configured with the documents and drive.file scopes, and yourself added as a test user while the app is in testing mode. Create a Desktop app OAuth client, download the JSON and save it in the project root as `credentials.json`. Node.js 18 or higher and Git; then `npm install` and `npm run build` to compile to `dist/`. The first run is manual and interactive: start the compiled server, open the printed Google URL, approve, and paste the code back — the browser will land on a "site can't be reached" error, which is expected, and the code is the value between `code=` and `&scope` in the address bar. That writes `token.json`, which grants access from then on without asking again — protect it like a password, and keep both files out of version control. MIT licensed.