Prakul95/mcp-server-google-docs

Google Docs MCP Server

A Model Context Protocol (MCP) server that lets AI assistants (e.g., Claude) work directly with your Google Docs. It can list Google Docs in Drive, fetch entire document content, generate quick previews, and create & edit Docs—all through MCP tools.

Perfect for drafting, summarizing, outlining, note‑taking, and collaborative edits without leaving chat.

✨ Features

• List Docs (Drive, Docs only) – Query your Drive for Google Docs (application/vnd.google-apps.document) with title, ID, and metadata. Supports folder scoping and extra Drive query parameters.
• Get full document – Retrieve the complete Google Docs structure (suitable for summarization, analysis, and formatting). Supports fetching tab content.
• Create Docs – Spin up a new Doc with an initial title and (optionally) place it inside a Drive folder.
• Update Docs – Apply edits to an existing Doc (insert/append text, headings/paragraph styles, replacements).
• Copy Docs - Creates a copy of an existing Doc.

🧰 Tools exposed by the server (exact names)

These tool names reflect your implementation:

• list-docs — List Google Docs from Drive. Can filter by folder ID and additional Drive query parameters.
• get-doc — Retrieve a Google Doc’s metadata and/or content. Supports fetching tab content.
• create-doc — Create a new blank Google Doc with a given title. Optionally place it in a Drive folder.
• update-doc — Update an existing Google Doc (e.g., insert/replace text, apply styles).
• copy-doc — copy an existing Google Doc optionally with a given title.

Demo Video

🧪 Example prompts (Claude / MCP client)

1. Get me a list of all the docs created yesterday with AI in their name.
2. Get the doc xyz and show details.
3. Get the same doc and show a preview of 400 chars.
4. Summarize the doc
5. Create a doc with name new_file_name
6. Add list of commands in this doc and write it in a fashion that explains the intent of this doc. Also add the current temp and lunar phase of the moon today.
7. Update this list with 5 random cities from Washington and add the temp of those cities as well.

🔐 Google Cloud Setup

1. Open Google Cloud Console and create/select a project.
2. Enable APIs for the project:
   • Google Docs API
   • Google Drive API
3. Create OAuth 2.0 credentials:
   • Type: Desktop app (recommended for local MCP clients).
   • Download the credentials JSON and save it locally (e.g., gcp-oauth.keys.json).
4. While the app is in Test mode, add your Google account under OAuth consent screen → Test users.

Recommended scopes (least‑privilege):

• Read only:
  • https://www.googleapis.com/auth/documents.readonly
  • https://www.googleapis.com/auth/drive.metadata.readonly
• Read/Write (create & edit):
  • https://www.googleapis.com/auth/documents
  • https://www.googleapis.com/auth/drive.file (or drive if you truly need broad access)

Tokens issued in Test mode often expire in ~7 days; you can re‑auth when prompted or publish the app to extend token lifetimes.

⚙️ Install & Run

Prerequisites

• Node.js LTS (v18+ recommended)
• npm or pnpm

Local install

git clone https://github.com/Prakul95/mcp-server-google-docs.git
cd mcp-server-google-docs
npm install
npm run build

Start the server (stdio MCP)

// TBA ypu can build the project and directluy update the claude_desktop_config.json, steps mentioned below

# From repo root
npm start             # if defined in package.json
# or explicitly:
node ./dist/server.js

🖥️ Configure your MCP client

Claude Desktop

Edit:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add an entry like:

{
  "mcpServers": {
    "google-docs": {
      "command": "node",
      "args": ["<absolute-path-to>/mcp-server-google-docs/dist/server.js"],
      "env": {
        "GOOGLE_OAUTH_CREDENTIALS": "<absolute-path-to>/gcp-oauth.keys.json"
      }
    }
  }
}

Tip: While iterating, you can wire up a dev command (e.g., tsx src/server.ts) if you prefer to run TypeScript directly.

Cursor / other MCP clients

Use the MCP configuration UI or a JSON config similar to the above (pointing to the built server). Ensure the GOOGLE_OAUTH_CREDENTIALS env var points to your credentials JSON.

🔑 Authentication flow

• On the first call, your MCP client will open a browser window for Google OAuth.
• Tokens are stored locally (location depends on your client/server setup).
• If you see auth errors, verify:
  • Your OAuth client is of type Desktop.
  • Your user is added under Test users (if app is in test mode).
  • GOOGLE_OAUTH_CREDENTIALS points to a valid credentials JSON file.

Re‑authentication (Test Mode)

• Expect token expiry roughly weekly in test mode.
• Re‑run your client and complete the OAuth prompt again.
• Publishing the app reduces re‑auth frequency (expect an “unverified app” banner).

🧪 Quick sanity checks

TBA

🛡️ Security notes

• Keep gcp-oauth.keys.json out of version control.
• Prefer least‑privilege scopes unless you truly need write access.
• Be mindful that logs may include tool inputs. Avoid logging sensitive data in production.

🧩 Implementation details (high level)

• Drive listing uses the Drive API with mimeType = 'application/vnd.google-apps.document' filters; combined with folder and custom Drive query.
• Full document fetch uses the Docs API documents.get to return structured content (paragraphs, tables, etc.).
• Create/update use Docs API documents.create and documents.batchUpdate with structured operations.

❓ Troubleshooting

1. Credentials file not found
   • Ensure GOOGLE_OAUTH_CREDENTIALS points to an absolute path.
2. “This app isn’t verified”
   • Expected for unpublished apps. You can proceed (or publish).
3. Auth keeps expiring
   • Normal in test mode; re‑auth or publish the app.
4. 403 / insufficient permissions
   • Add missing scopes (e.g., documents for write, drive.file for creating files).
5. Docs list is empty
   • Check your folder filter / Drive query and confirm the account has Docs in Drive.

📄 License

MIT

🙌 Acknowledgements

• Setup flow inspired by community MCP servers (e.g., Google Calendar variants); adapted here for Google Docs tooling.