Gmail MCP Connector
A Gmail MCP server running on Cloudflare Workers that enables reading, searching, labeling, drafting, sending, and managing Gmail messages, including fetching raw attachment bytes, with per-user OAuth authorization.
README
Gmail MCP Connector
A full Gmail MCP connector running as a Cloudflare Worker. It reads, searches, labels, drafts, sends, and manages Gmail — including minting short-lived download URLs for attachments so the raw bytes can be fetched out-of-band without ever entering the model's context, a key capability this connector adds over generic Gmail integrations.
Authorization follows the MCP authorization spec (OAuth 2.1): each user who connects runs the OAuth handshake themselves, and the client holds the token — this connector is multi-user and does not use a shared account secret.
How auth works
This Worker plays two OAuth roles at once (via @cloudflare/workers-oauth-provider):
- To the MCP client, it is the authorization server + resource server. It issues its own token to the client and validates that token on every
/api/mcprequest. - To Google, it is an OAuth client. It bridges the user to Google's consent screen and obtains a separate Google token.
Per the spec, the client's token is never passed through to Google. Each user's Google tokens are stored encrypted in KV (in the grant props) and used to call Gmail on their behalf. The client only ever holds the Worker's token.
MCP client ──token(A)──▶ Worker (RS/AS) ──token(B)──▶ Google/Gmail
◀─issues A── ◀──issues B──
token A never forwarded to Google
Endpoints
| Endpoint | Purpose |
|---|---|
POST /api/mcp |
MCP JSON-RPC endpoint (requires a valid access token) |
GET /attachment?ticket=<id> |
Redeem a one-time attachment download ticket; streams the raw bytes (no auth header — the ticket is the credential) |
GET /authorize |
Authorization UI — bridges the user to Google consent |
GET /oauth/callback/google |
Google redirect target; completes the grant |
POST /token, POST /register |
Token + dynamic client registration (library-served) |
GET /.well-known/oauth-protected-resource |
RFC 9728 metadata (library-served) |
GET /.well-known/oauth-authorization-server |
RFC 8414 metadata (library-served) |
Setup
-
Create a KV namespace for token storage and put its id in
wrangler.toml:wrangler kv namespace create OAUTH_KV # paste the returned id into the [[kv_namespaces]] block in wrangler.toml -
Create a Google OAuth client (type: Web application) in the Google Cloud console, and enable the Gmail API. Add this authorized redirect URI:
{your-worker-origin}/oauth/callback/google(e.g.
https://gmail-mcp.your-subdomain.workers.dev/oauth/callback/google, or your custom domain.) -
Set the client secrets:
wrangler secret put GOOGLE_CLIENT_ID wrangler secret put GOOGLE_CLIENT_SECRETNo
GOOGLE_REFRESH_TOKEN— tokens are now obtained per-user at connect time. -
(Optional) Set
BASE_URLinwrangler.tomlif you serve behind a custom domain. If left empty, the Worker derives redirect URIs from the incoming request origin. -
Deploy:
wrangler deploy -
Connect a client. Add
{your-worker-origin}/api/mcpas a custom MCP server. The client performs dynamic client registration + the OAuth flow automatically; you'll be sent to Google to sign in and consent on first connect.
Tools
search_threads, get_thread, get_message, get_attachment, send_email, reply_to_thread, create_draft, list_drafts, trash_message, list_labels, create_label, label_message, unlabel_message, label_thread, unlabel_thread.
Attachments
get_attachment does not return bytes — it returns a short-lived download URL:
{ "download_url": "https://<worker>/attachment?ticket=<id>", "method": "GET", "expires_in_seconds": 600 }
Do a plain HTTP GET on download_url to stream the raw bytes. No auth header is needed — the opaque, high-entropy ticket in the URL is the credential. The ticket is one-time use and expires after 10 minutes.
This keeps the attachment bytes out of the tool result (and therefore out of the model's context): the consuming agent fetches them out-of-band into its own runtime, so large attachments that would blow the MCP tool-result size cap are no longer a problem. Under the hood, minting the URL writes a ticket into OAUTH_KV carrying that grant's Google access token; the GET /attachment endpoint redeems it server-side, fetches the bytes from Gmail, and deletes the ticket. The Google access token never appears in a URL, a tool result, or logs.
If the ticket's copied access token has expired by download time (rare within the 10-minute window), the endpoint returns 401 — just call get_attachment again for a fresh URL.
Optionally pass file_name and mime_type (from get_message's attachment metadata) so the download response sets a correct Content-Type and Content-Disposition.
Development
pnpm install
./node_modules/.bin/tsc --noEmit # typecheck
wrangler dev # local dev server
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.