About google-docs
An MCP server for Google Docs that supports OAuth2 for personal accounts and service accounts for Google Workspace. It can read, create, edit, export, and share documents, with tool groups for access control. It suits individuals and organizations that want AI assistants to work with Google Docs.
Key features
- 01Read document body and heading outline
- 02Create documents with initial content
- 03Append, insert, and replace text
- 04Export to multiple file formats
- 05Tool group access control
- 06Read document comments
Use cases
- 01Getting a heading outline of a long document
- 02Exporting a document to Markdown or PDF
- 03Replacing a term across a document
- 04Reviewing comments in read-only mode
MCP support
This project is an MCP server that exposes document read, edit, export, comment, and sharing tools to MCP clients. Available tools can be limited through the GOOGLE_DOCS_ENABLED_TOOLGROUPS setting.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
npx -y google-docs-workspace-mcp-serverGOOGLE_DOCS_OAUTH_CLIENT_IDSecretOAuth2 client ID from Google Cloud Console.GOOGLE_DOCS_OAUTH_CLIENT_SECRETSecretOAuth2 client secret from Google Cloud Console.GOOGLE_DOCS_OAUTH_REFRESH_TOKENSecretLong-lived refresh token obtained from the oauth-setup command.GOOGLE_DOCS_SERVICE_ACCOUNT_CLIENT_EMAILService account email address for domain-wide delegation.GOOGLE_DOCS_SERVICE_ACCOUNT_PRIVATE_KEYSecretService account private key in PEM format.GOOGLE_DOCS_IMPERSONATE_EMAILEmail address to impersonate in Google Workspace.GOOGLE_DOCS_ENABLED_TOOLGROUPSComma-separated list of tool groups to enable (readonly, readwrite, readwrite_external).Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
Tools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project READMEGoogle Docs Workspace MCP Server
An MCP (Model Context Protocol) server that provides Google Docs integration for AI assistants. Supports both personal Google accounts (via OAuth2) and Google Workspace accounts (via service account with domain-wide delegation).
Features
- Read documents: Fetch full body or just the heading outline of a Google Doc, by document ID or shareable URL
- Create documents: Create new Google Docs with optional initial content
- Edit documents:
- Append text to the end of a document
- Insert text at a specific index
- Replace all occurrences of a string
- Apply raw
documents.batchUpdaterequests for full structural control
- Delete documents: Move to Drive trash by default, or permanently delete with an explicit flag
- Export documents: Export to Markdown, HTML, plain text, PDF, DOCX, ODT, RTF, or EPUB
- Share documents (gated): Grant read/comment/write access to a user, group, domain, or anyone with a link
- Two auth methods: OAuth2 for personal accounts, service account for Google Workspace
- Permission gating: Restrict tools by group via
GOOGLE_DOCS_ENABLED_TOOLGROUPS
Installation
npm install google-docs-workspace-mcp-server
Or run directly with npx:
npx google-docs-workspace-mcp-server
Authentication
This server supports two authentication modes. Choose the one that matches your account type.
Option 1: OAuth2 (Personal Google Accounts)
Use this for personal @gmail.com accounts or any Google account without Workspace admin access.
Prerequisites
- Go to Google Cloud Console
- Create or select a project
- Enable the Google Docs API and the Google Drive API
- Configure the OAuth consent screen:
- Choose External user type
- Add yourself as a test user
- Publish the app (click "Publish" on the consent screen) to prevent refresh tokens from expiring every 7 days. No full Google verification is needed for personal use.
- Create OAuth 2.0 credentials:
- Choose Desktop app as the application type
- Important: Create new credentials after publishing the app
- Copy the Client ID and Client Secret
Getting a Refresh Token
Run the built-in setup command:
npx google-docs-workspace-mcp-server oauth-setup <client_id> <client_secret>
You can also pass credentials via environment variables:
GOOGLE_DOCS_OAUTH_CLIENT_ID=... GOOGLE_DOCS_OAUTH_CLIENT_SECRET=... npx google-docs-workspace-mcp-server oauth-setup
Port conflict? If port 3000 is already in use, specify a different port:
PORT=3001 npx google-docs-workspace-mcp-server oauth-setup <client_id> <client_secret>
Desktop app credentials automatically allow http://localhost redirects on any port, so no additional Google Cloud Console configuration is needed.
This will open your browser for Google sign-in and print the refresh token. You only need to do this once — the refresh token does not expire (as long as the OAuth consent screen is published).
Environment Variables (OAuth2)
| Variable | Required | Description |
|---|---|---|
GOOGLE_DOCS_OAUTH_CLIENT_ID | Yes | OAuth2 client ID from Google Cloud Console |
GOOGLE_DOCS_OAUTH_CLIENT_SECRET | Yes | OAuth2 client secret |
GOOGLE_DOCS_OAUTH_REFRESH_TOKEN | Yes | Refresh token from the setup script |
GOOGLE_DOCS_ENABLED_TOOLGROUPS | No | Comma-separated list of tool groups (default: all) |
Claude Desktop Configuration (OAuth2)
{
"mcpServers": {
"google-docs": {
"command": "npx",
"args": ["google-docs-workspace-mcp-server"],
"env": {
"GOOGLE_DOCS_OAUTH_CLIENT_ID": "your-client-id.apps.googleusercontent.com",
"GOOGLE_DOCS_OAUTH_CLIENT_SECRET": "your-client-secret",
"GOOGLE_DOCS_OAUTH_REFRESH_TOKEN": "your-refresh-token"
}
}
}
}
Option 2: Service Account (Google Workspace)
Use this for Google Workspace organizations where a domain admin can grant domain-wide delegation.
Prerequisites
- Go to Google Cloud Console
- Create or select a project
- Enable the Google Docs API and the Google Drive API
- Create a service account with domain-wide delegation enabled
- In Google Workspace Admin Console, grant the service account access to the following scopes:
https://www.googleapis.com/auth/documents(read and edit documents)https://www.googleapis.com/auth/drive(full read/write access to the impersonated user's Drive — required to export, read comments on, and manage arbitrary existing docs, including files this app did not create)
- Download the JSON key file
Environment Variables (Service Account)
| Variable | Required | Description |
|---|---|---|
GOOGLE_DOCS_SERVICE_ACCOUNT_CLIENT_EMAIL | Yes | Service account email address |
GOOGLE_DOCS_SERVICE_ACCOUNT_PRIVATE_KEY | Yes | Service account private key (PEM format) |
GOOGLE_DOCS_IMPERSONATE_EMAIL | Yes | Email address to impersonate |
GOOGLE_DOCS_ENABLED_TOOLGROUPS | No | Comma-separated list of tool groups to enable (default: all groups) |
You can find the client_email and private_key values in your service account JSON key file.
Claude Desktop Configuration (Service Account)
{
"mcpServers": {
"google-docs": {
"command": "npx",
"args": ["google-docs-workspace-mcp-server"],
"env": {
"GOOGLE_DOCS_SERVICE_ACCOUNT_CLIENT_EMAIL": "[email protected]",
"GOOGLE_DOCS_SERVICE_ACCOUNT_PRIVATE_KEY": "-----BEGIN PRIVATE KEY-----\nMIIE...\n-----END PRIVATE KEY-----",
"GOOGLE_DOCS_IMPERSONATE_EMAIL": "[email protected]"
}
}
}
}
Note: For the private key, you can either:
- Use the key directly with
\nfor newlines (as shown above) - Set the environment variable from a shell that preserves newlines
Tool Groups
The server supports three tool groups for permission-based access control:
| Group | Tools Included | Risk Level |
|---|---|---|
readonly | get_document, get_document_outline, export_document, list_comments | Low |
readwrite | All readonly tools + create_document, update_document, delete_document, append_text, insert_text, replace_text | Medium |
readwrite_external | All readwrite tools + share_document | High |
By default, all tool groups are enabled. To restrict access, set the GOOGLE_DOCS_ENABLED_TOOLGROUPS environment variable:
# Read-only access (no writes or sharing)
GOOGLE_DOCS_ENABLED_TOOLGROUPS=readonly
# Read and write, but no sharing
GOOGLE_DOCS_ENABLED_TOOLGROUPS=readwrite
# Full access including sharing (default)
GOOGLE_DOCS_ENABLED_TOOLGROUPS=readwrite_external
Security Note: The share_document tool is in a separate readwrite_external group because it can grant access to users outside your organization, which carries higher risk than purely internal edits.
Available Tools
All tools that take a document_id accept either a raw Google Docs ID (e.g., 1AbCdEf...) or a shareable Docs URL (e.g., https://docs.google.com/document/d/1AbCdEf.../edit). The server extracts the ID automatically.
get_document
Fetch a Google Doc and return its title, outline, and full body.
Parameters:
document_id(string, required): Google Docs ID or URLformat(string, optional):text(default) for human-readable output, orjsonfor the rawdocuments.getAPI resource
get_document_outline
Return only the heading structure of a document — useful for getting your bearings on a long doc without pulling the full body.
Parameters:
document_id(string, required): Google Docs ID or URL
create_document
Create a new Google Doc.
Parameters:
title(string, required): Document titleinitial_content(string, optional): Plain text inserted into the new document
update_document
Apply raw documents.batchUpdate requests. This is the escape hatch for any Docs API operation not covered by the more focused tools (e.g., applying paragraph styles, inserting tables, manipulating named ranges).
Parameters:
document_id(string, required): Google Docs ID or URLrequests(array, required): Array ofdocuments.batchUpdaterequest objects (API reference)
Example:
{
"document_id": "1AbCdEf...",
"requests": [
{
"insertText": {
"location": { "index": 1 },
"text": "Hello, world!\n"
}
},
{
"updateParagraphStyle": {
"range": { "startIndex": 1, "endIndex": 14 },
"paragraphStyle": { "namedStyleType": "HEADING_1" },
"fields": "namedStyleType"
}
}
]
}
append_text
Append text to the end of a document. Resolves the end-of-body index for you.
Parameters:
document_id(string, required): Google Docs ID or URLtext(string, required): Text to append
insert_text
Insert text at a specific index.
Parameters:
document_id(string, required): Google Docs ID or URLindex(number, required): Insertion index (must be ≥ 1)text(string, required): Text to insert
replace_text
Find and replace text throughout a document. Reports the number of occurrences replaced.
Parameters:
document_id(string, required): Google Docs ID or URLfind(string, required): String to findreplace(string, required): Replacement stringmatch_case(boolean, optional): Whether matching is case-sensitive (default:false)
delete_document
Move a document to Drive trash by default, or permanently delete it.
Parameters:
document_id(string, required): Google Docs ID or URLpermanent(boolean, optional): Whentrue, permanently delete instead of trashing (default:false)
export_document
Export a document in a different format.
Parameters:
document_id(string, required): Google Docs ID or URLformat(string, required): One ofmarkdown,html,txt,pdf,docx,odt,rtf,epub
Text formats (markdown, html, txt) are returned inline. Binary formats (pdf, docx, odt, rtf, epub) are returned base64-encoded.
list_comments
List the comments on a document, including threaded replies and resolved state. Reads the Drive comments.list API and surfaces each comment's author, timestamps, resolved/open state, the anchored quoted text it refers to, the comment body, and any replies (including replies that resolved or reopened the thread).
Parameters:
document_id(string, required): Google Docs ID or URLinclude_resolved(boolean, optional): Include resolved (closed) threads (default:true). Setfalseto see only open comments.include_deleted(boolean, optional): Include deleted comments (default:false)
Reading comments on a document this app did not create requires the https://www.googleapis.com/auth/drive scope (the narrower drive.file scope alone is insufficient).
share_document
Grant access to a user, group, domain, or anyone with the link. Gated behind the readwrite_external group.
Parameters:
document_id(string, required): Google Docs ID or URLtype(string, required): One ofuser,group,domain,anyonerole(string, required): One ofreader,commenter,writeremail_address(string, conditional): Required fortype=userortype=groupdomain(string, conditional): Required fortype=domainsend_notification_email(boolean, optional): Whether Google sends an email notification (default:true)
Development
Setup
# Install dependencies
npm run install-all
# Build
npm run build
# Run in development mode
npm run dev
Testing
This server uses a four-tier test strategy:
| Tier | Command | Hits real APIs? | Credentials needed |
|---|---|---|---|
| Functional | npm test | No | None |
| Integration | npm run test:integration | No | None |
| E2E | npm run test:e2e | Yes | OAuth2 or service account (encrypted) |
| Manual | npm run test:manual | Yes | OAuth2 or service account (your .env) |
The functional and integration tests run in CI on every commit. The e2e tests run against the live Google Docs and Drive APIs using a pinned test document and a refresh token committed (encrypted) to the repo.
Running E2E Tests
E2e tests live in tests/e2e/ and exercise the full MCP server against the real Google APIs. They include:
- Read-only assertions against a pinned, pre-existing test document
- A self-cleaning lifecycle test that creates → mutates → exports → shares → deletes a fresh document end-to-end
- An error-handling check for non-existent document IDs
The full suite runs once per auth mode — both OAuth2 (personal account) and Service Account (Workspace impersonation) are always exercised when the suite is configured. This catches auth-mode-specific bugs (token refresh vs JWT signing, impersonation, the SA-specific subject claim) before they reach production. The suite is fail-fast on partial credentials: if any OAuth or SA env var is set, all required vars must be set — otherwise the suite throws at module load. The only valid "no creds" state is fork PRs where MCP_SERVERS_MASTER_KEY is unavailable, in which case the entire suite skips.
Credentials are stored encrypted in tests/e2e/.env.enc and decrypted at runtime using the MCP_SERVERS_MASTER_KEY shared by the entire mcp-servers/ workspace.
To run e2e tests locally:
# 1. Decrypt the e2e credentials (one-time per checkout)
cd mcp-servers
MCP_SERVERS_MASTER_KEY="<key>" ./scripts/decrypt-credentials.sh google-docs
# 2. Run the suite (this also rebuilds the server and the test client)
cd servers/google-docs
npm run test:e2e
The master key is in 1Password under "MCP Servers Master Key" — ask a teammate with access if you don't have it. The decrypted .env is gitignored; only .env.enc is committed.
If tests/e2e/.env is missing or the credentials are invalid, the suite gracefully skips itself rather than failing — so CI on forks (where the master key isn't available) won't break.
For details on the pinned test document, OAuth scopes, token rotation, the SA provisioning steps, and the full list of tools exercised, see tests/e2e/STATE.md.
Updating E2E Credentials
If the refresh token gets revoked, expires, or the OAuth client changes, mint a new refresh token and re-encrypt:
# 1. Mint a new refresh token via the OAuth flow (see tests/e2e/STATE.md for the full procedure)
# 2. Update tests/e2e/.env with the new GOOGLE_DOCS_OAUTH_REFRESH_TOKEN
# 3. Re-encrypt
cd mcp-servers
MCP_SERVERS_MASTER_KEY="<key>" ./scripts/encrypt-credentials.sh google-docs
# 4. Commit the updated .env.enc
git add servers/google-docs/tests/e2e/.env.enc
License
MIT
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.