Google Docs Preview MCP
Enables creating anchored comments, replies, and real suggested edits in Google Docs from Claude.
README
<!--ai-->
Google Docs Preview MCP
Create anchored comments, replies, and real suggested edits in Google Docs from Claude.
Google’s hosted Docs MCP recognizes the preview operations but does not expose enough of the Docs API schema to use them:
| Operation | Hosted MCP problem | This implementation |
|---|---|---|
| Create comments | Strips insertComment.content |
Passes the complete request |
| Suggest edits | Omits top-level writeControl |
Exposes writeControl.writeMode: SUGGEST |
| Recover preview objects | Omits comment-view parameters | Exposes comment, suggestion, and tab view controls |
The underlying Google Docs REST API accepts all three operations. See what failed and why.
Choose a setup
| Setup | Works in | Trust boundary | Status |
|---|---|---|---|
| Local stdio | Claude Desktop and Claude Code | Your computer, Google, and Claude | Recommended for sensitive documents; tested |
| Organization connector | Claude.ai and other cloud clients | Adds the organization’s hosting and storage providers | End-to-end pilot passed; production ownership and security review remain |
| Personal remote | Claude.ai and other cloud clients | Adds a developer-owned host | Test only; do not use for sensitive documents |
The local server is the current recommendation. Claude.ai cannot connect to localhost or stdio; use Claude Desktop or Claude Code for the local option. Tool results still go to Claude for processing.
The local token file is plaintext JSON protected by Unix file permissions, not an OS keychain. Local execution removes the independent MCP host, but it does not protect against malware running as your user or an administrator.
Local quick start
npm test
npm run auth -- --client /path/to/desktop-oauth-client.json
claude mcp add --scope user google-docs-preview-local -- \
node "$PWD/bin/docs-mcp-local.js"
The OAuth flow requests:
https://www.googleapis.com/auth/documents
This is the minimum scope for arbitrary document IDs, but Google classifies it as Sensitive and describes it as permission to see, edit, create, and delete all Google Docs documents. A drive.file design can narrow access to files a user explicitly selects, but it requires a Google Picker or per-file grant flow that this document-ID-only prototype does not implement.
See the complete setup guide before authorizing an account.
Organization connector
One Google Cloud project and OAuth app can serve the team. Staff do not create Cloud projects or handle credentials: a Claude Owner publishes the connector once, then each registered Preview tester clicks Connect and authorizes their own Google account. Read the staff onboarding guide, administrator runbook, and pilot result.
The pilot deployment is developer-owned. Move the deployment, Redis store, secrets and administration to 80,000 Hours accounts before approving confidential documents.
Repository map
api/ Vercel HTTP transport
bin/ Local authorization and stdio entrypoints
src/ Shared MCP and Google Docs logic
test/ Protocol, passthrough, and token-refresh tests
docs/ Setup, security analysis, evidence, and guide images
archive/ Superseded documents retained for context
docs/reproduce/ is the concise maintained path from decisive prompts through
method, scripts, inputs, and checks to the current output.
Verification
npm test
The test suite covers the preview schema, complete request forwarding, Docs-only scope metadata, canonical remote origins, redirect rejection, local token refresh, file permissions, OAuth challenges, and stateless protocol methods. Claude.ai, Claude Code and Claude Desktop have each passed live reads. The Claude.ai organization pilot created and recovered an anchored comment, its reply and a real unresolved suggestion, then the Google Docs UI was inspected.
Security status
- Local tokens are plaintext JSON stored outside the repository with mode
600. - Docs tool calls construct requests only for the fixed
docs.googleapis.comorigin; local OAuth separately contacts Google’s documented authorization and token endpoints. - Tool annotations mark writes as destructive, but approval behavior is enforced by the MCP client. Configure per-call approval for writes.
- The remote uses separate MCP and Google tokens. Google grants are encrypted at rest; Claude never receives the Google refresh token.
- The pilot proved one member’s flow. A second user and organizational unit have not been tested.
- The pilot host and Redis resources are still developer-owned. Organization ownership, admin-led offboarding, rate limits, monitoring and security review remain.
Read SECURITY.md, the organization-owned deployment decision, and the full threat model. <!--/ai-->
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
Neon Database
MCP server for interacting with Neon Management API and databases
E2B
Using MCP to run code via e2b.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.