google-workspace-mcp-sidecar
A policy-enforcing MCP sidecar that provides Gmail and Google Calendar tools, enabling email drafting, thread management, calendar event retrieval, and time suggestions with server-side recipient and visibility policies.
README
google-workspace-mcp-sidecar
A Node.js Docker service that acts as a policy-enforcing MCP sidecar for Gmail and Google Calendar.
- Exposes a local Gmail MCP endpoint and a local Calendar MCP endpoint
- Translates MCP tool calls into direct Gmail API and Calendar API requests
- Enforces Gmail recipient policy server-side and Gmail visibility-cutoff policy on thread reads
- Keeps OAuth credentials and tokens inside the sidecar only
Requirements
- Node.js 24+
- npm 10+
- Docker (for containerized deployment)
- A Google Cloud project with Gmail API and Calendar API enabled
Install dependencies
npm install
Run the test suite
npm test
Type-check
npm run typecheck
Lint and format
npm run lint
npm run fmt:check # check only
npm run fmt # format in place
Build
Compile TypeScript
npm run build
Build the Docker image
docker build -t google-workspace-mcp-sidecar:latest .
Publish the Docker image
docker tag google-workspace-mcp-sidecar:latest <your-registry>/google-workspace-mcp-sidecar:latest
docker push <your-registry>/google-workspace-mcp-sidecar:latest
Configuration
The service loads a single JSON config file. The path defaults to ./config.json and can be
overridden with the CONFIG_PATH environment variable.
Secret values (OAuth client ID and secret) may be provided via environment variable interpolation
using ${ENV_VAR_NAME} syntax in the config file.
Example config
{
"listen_addr": "0.0.0.0:8080",
"oauth_callback_url": "https://example.com/oauth/callback",
"gmail": {
"local_path": "/gmail/mcp",
"visibility_cutoff_date": "2026-01-01",
"allowed_recipients": ["*"]
},
"calendar": {
"local_path": "/calendar/mcp"
},
"oauth": {
"client_id": "${GOOGLE_OAUTH_CLIENT_ID}",
"client_secret": "${GOOGLE_OAUTH_CLIENT_SECRET}",
"token_dir": "/state/tokens",
"allowed_account_emails": ["owner@example.com"]
}
}
Config fields
| Field | Description |
|---|---|
listen_addr |
HTTP listen address and port, e.g. 0.0.0.0:8080 |
oauth_callback_url |
Full public URL of the OAuth callback, served on the same port |
gmail.local_path |
URL path prefix for the Gmail MCP endpoint |
gmail.visibility_cutoff_date |
Oldest visible Gmail date for thread reads in YYYY-MM-DD format |
gmail.allowed_recipients |
Allowlist of recipient addresses. Use ["*"] to allow all |
calendar.local_path |
URL path prefix for the Calendar MCP endpoint |
oauth.client_id |
Google OAuth client ID |
oauth.client_secret |
Google OAuth client secret |
oauth.token_dir |
Directory where the token file is persisted |
oauth.allowed_account_emails |
Google account emails allowed to complete first-run OAuth |
Google Cloud setup
- Enable the Gmail API and Calendar API in your Google Cloud project
- Configure an OAuth consent screen
- Create an OAuth 2.0 client (type: Web application)
- Add your
oauth_callback_urlas an authorized redirect URI
First-run authentication
If no valid token exists when the service starts, it prints the Google authorization URL to the terminal and listens for the OAuth callback on the same port as the MCP endpoints. Open the URL in a browser, complete the Google sign-in, and the service will persist the token and begin serving automatically. No separate script is needed.
Run the service locally
CONFIG_PATH=./config.json npm start
Run with Docker
On first run (no token present) the container prints the Google auth URL and waits for the callback on the configured port. Complete the browser flow and the container continues startup automatically.
docker run -d \
--name google-workspace-mcp-sidecar \
-e GOOGLE_OAUTH_CLIENT_ID=your-client-id \
-e GOOGLE_OAUTH_CLIENT_SECRET=your-client-secret \
-v $(pwd)/config.json:/app/config.json:ro \
-v $(pwd)/state:/state \
-p 8080:8080 \
google-workspace-mcp-sidecar:latest
Follow the logs to get the authorization URL: docker logs -f google-workspace-mcp-sidecar.
To select the runtime user, add --user <uid>:<gid> to docker run:
docker run -d \
--name google-workspace-mcp-sidecar \
--user 1000:1000 \
-e GOOGLE_OAUTH_CLIENT_ID=your-client-id \
-e GOOGLE_OAUTH_CLIENT_SECRET=your-client-secret \
-v $(pwd)/config.json:/app/config.json:ro \
-v $(pwd)/state:/state \
-p 8080:8080 \
google-workspace-mcp-sidecar:latest
Steady-state serving
Once a token is present, start the sidecar:
docker run -d \
--name google-workspace-mcp-sidecar \
-e GOOGLE_OAUTH_CLIENT_ID=your-client-id \
-e GOOGLE_OAUTH_CLIENT_SECRET=your-client-secret \
-v $(pwd)/config.json:/app/config.json:ro \
-v $(pwd)/state:/state \
-p 8080:8080 \
google-workspace-mcp-sidecar:latest
The service will reuse the persisted token from /state/tokens/token.json and refresh it
automatically without requiring a new manual login.
Example docker-compose.yml
services:
google-workspace-mcp-sidecar:
image: google-workspace-mcp-sidecar:latest
restart: unless-stopped
user: "1000:1000"
environment:
GOOGLE_OAUTH_CLIENT_ID: "${GOOGLE_OAUTH_CLIENT_ID}"
GOOGLE_OAUTH_CLIENT_SECRET: "${GOOGLE_OAUTH_CLIENT_SECRET}"
volumes:
- ./config.json:/app/config.json:ro
- ./state:/state
ports:
- "8080:8080"
MCP endpoint URLs
| Endpoint | URL |
|---|---|
| Gmail MCP | http://<host>:8080/gmail/mcp |
| Calendar MCP | http://<host>:8080/calendar/mcp |
Security
This service is a sidecar, not a general-purpose internet-facing API.
Its security model is intentionally narrow:
- the sidecar owns Google OAuth credentials and tokens
- Gmail recipient policy is enforced server-side for
create_draft - Gmail visibility cutoff is enforced on thread reads
- secret-bearing files are expected to be locked down on disk
What this service intentionally does not do
The sidecar does not try to solve every access-control problem itself.
- It does not provide a full user/session/authentication layer for downstream MCP clients.
- It does not try to be safe to expose broadly to browsers or untrusted networks.
- It does not replace network-level controls, reverse-proxy policy, VPN boundaries, or firewalling.
That is intentional. The intended deployment model is a trusted local or private sidecar sitting near the MCP client that is allowed to use the authenticated Google account.
Public exposure warning
Do not expose the Gmail or Calendar MCP endpoints directly to the public internet.
Insecure deployment patterns include:
- binding the service to a public interface and publishing
/gmail/mcpor/calendar/mcp - putting the MCP endpoints behind a public reverse proxy without additional request filtering
- allowing arbitrary browsers, web pages, or untrusted clients to reach the MCP endpoints
If you expose the MCP endpoints carelessly, an attacker may be able to use the sidecar as a confused deputy against the authenticated Google account.
Only the callback should be public
The only route that is meant to be reachable from the public internet is the OAuth callback path configured by oauth_callback_url.
That route exists only to complete the Google OAuth redirect during bootstrap. The Gmail MCP endpoint and Calendar MCP endpoint should remain private and reachable only by the trusted MCP client or a tightly controlled private network path.
Recommended deployment posture:
- expose only the callback path externally
- keep
/gmail/mcpand/calendar/mcpon localhost, a private container network, or behind strict network allowlists - treat the MCP endpoints as privileged local/private control surfaces
MCP client configuration example
{
"mcpServers": {
"gmail": {
"url": "http://google-workspace-mcp-sidecar:8080/gmail/mcp"
},
"calendar": {
"url": "http://google-workspace-mcp-sidecar:8080/calendar/mcp"
}
}
}
Available MCP tools
Gmail tools
| Tool | Description |
|---|---|
search_threads |
Search Gmail threads |
get_thread |
Fetch a thread by ID |
list_drafts |
List drafts |
create_draft |
Create a new draft |
create_draft also accepts optional thread_id, in_reply_to, and references fields for
threaded replies. When thread_id is provided without the reply headers, the sidecar derives them
from the target Gmail thread before creating the draft.
Known limitations
list_draftsdoes not enforcevisibility_cutoff_date. This is an accepted exception: drafts may expose pre-cutoffid,snippet, andinternalDatevalues by design.respond_to_eventupdates the attendee list with a read-then-patch flow. Concurrent attendee changes can be overwritten if another client updates the same event between those two calls.
Calendar tools
| Tool | Description |
|---|---|
list_calendars |
List accessible calendars |
list_events |
List events in a calendar |
get_event |
Fetch a single event |
suggest_time |
Find an available time slot |
create_event |
Create a new event |
update_event |
Update an existing event |
respond_to_event |
Accept, decline, or tentatively accept an invitation |
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.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
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.