gcal-mcp-remote-ldraney 0.1.1

Remote MCP connector wrapping gcal-mcp with Google OAuth 2.0 over Streamable HTTP

gcal-mcp-remote

Remote MCP server wrapping gcal-mcp with Google OAuth 2.0 and Streamable HTTP transport — designed for Claude.ai connectors.

How it works

Claude.ai ──HTTP+Bearer──> gcal-mcp-remote ──Google API──> Google Calendar
                                │
                          imports gcal-mcp's FastMCP (all 14 tools)
                          patches get_client() with per-request ContextVar
                          adds Google OAuth + /health + /oauth/callback

Three-party OAuth: Claude ↔ this server ↔ Google. Each user authorizes their own Google Calendar via OAuth. The server stores per-user Google refresh tokens (encrypted at rest).

Prerequisites

1. A Web application OAuth client in Google Cloud Console:
   • Create Credentials → OAuth client ID → Type: Web application
   • Add authorized redirect URI: {BASE_URL}/oauth/callback
   • Note the Client ID and Client Secret
2. Google Calendar API enabled in the same project

Setup

git clone https://github.com/ldraney/gcal-mcp-remote.git
cd gcal-mcp-remote
cp .env.example .env
# Edit .env with your Google OAuth credentials, BASE_URL, and SESSION_SECRET
make install
make run

Environment Variables

Variable	Description
GCAL_OAUTH_CLIENT_ID	Google OAuth Web application client ID
GCAL_OAUTH_CLIENT_SECRET	Google OAuth Web application client secret
SESSION_SECRET	Random secret for encrypting token store
BASE_URL	Public HTTPS URL where this server is reachable
HOST	Bind address (default: 127.0.0.1)
PORT	Listen port (default: 8001)

Verification

curl http://127.0.0.1:8001/health                                    # {"status": "ok"}
curl http://127.0.0.1:8001/.well-known/oauth-authorization-server    # OAuth metadata
curl -X POST http://127.0.0.1:8001/mcp                               # 401 (auth required)

Deploying

Standalone

Use any HTTPS tunnel (Tailscale Funnel, ngrok, Cloudflare Tunnel) to expose the server publicly, then add the URL as a Claude.ai connector.

A systemd service file is provided in systemd/gcal-mcp-remote.service.

Kubernetes (production)

Production deployment is managed by mcp-gateway-k8s, which runs this server as a pod with Tailscale Funnel ingress. That repo contains the Dockerfile, K8s manifests, and secrets management.

Full integration testing (Claude.ai connector → OAuth → Google Calendar) is tracked in mcp-gateway-k8s#12 and #6.

License

MIT