The thoughtful part is the format handling: a model asking for a Google Doc gets Markdown, a Sheet gets CSV — text it can actually reason over instead of a raw export. Access is read-only by design, scoped to `drive.readonly`, so it can find and read but never change your files. The setup is the standard Google OAuth chore: a Cloud project, Desktop credentials, and a one-time browser auth to mint the token.
An MCP server for Google Drive that lets an AI model search, list and read files. It handles Google Workspace formats by converting them on read — Docs to Markdown, Sheets to CSV, Slides to plain text — so what comes back is usable text rather than an export blob.
- Files found by full-text search, returned with name, MIME type, file ID, last-modified time and size — `gdrive_search`
- A file's contents read by its Google Drive file ID — `gdrive_read_file`
- Google Workspace files converted on read — Docs to Markdown, Sheets to CSV, Slides to plain text, Drawings to PNG — with text and JSON as UTF-8 and other types base64-encoded
- Read-only access to Drive
A Google Cloud project with the Google Drive API enabled and OAuth 2.0 Desktop-app credentials, using the `drive.readonly` scope. The downloaded OAuth JSON goes into a `credentials/` directory as `gcp-oauth.keys.json`; running `node dist/index.js auth` completes the browser OAuth flow and saves the token beside it. Node.js v16 or higher; install is a clone plus `npm install` and `npm run build`. The package.json names it `@modelcontextprotocol/server-gdrive` at 0.6.2.
Build from source — clone the repository and build it, then point your client at the binary