Deployment guide

Three deployment models: Claude Code (stdio), container (HTTP), and Kubernetes.

Claude Code (stdio)

Add to .claude/settings.json:

{
  "mcpServers": {
    "google-workspace": {
      "command": "/path/to/mcp-google-workspace",
      "args": ["--policy", "/path/to/policy.json", "--eager-tools"]
    }
  }
}

Claude Code starts the binary as a child process, communicates over stdin/stdout, and shuts it down on exit. The --eager-tools flag loads all 53 helper tools at startup. Without it, only gws_discover and gws_batch are available initially.

To load MCP prompts (workflow recipes), add --prompts-dir:

{
  "mcpServers": {
    "google-workspace": {
      "command": "/path/to/mcp-google-workspace",
      "args": ["--policy", "/path/to/policy.json", "--eager-tools", "--prompts-dir", "/path/to/prompts"]
    }
  }
}

Container (HTTP)

podman run -p 3000:3000 \
  -v ./policy.json:/etc/mcp-google-workspace/policy.json:ro,Z \
  -v ./credentials.json:/etc/mcp-google-workspace/credentials.json:ro,Z \
  ghcr.io/fabiendupont/mcp-google-workspace:0.7.0

On Fedora and RHEL with SELinux, the :Z flag is required for bind mounts.

The image is FROM scratch -- about 6 MB, no shell, no OS packages. Available for linux/amd64 and linux/arm64.

To enable eager tools and prompts in the container, pass additional arguments:

podman run -p 3000:3000 \
  -v ./policy.json:/etc/mcp-google-workspace/policy.json:ro,Z \
  -v ./credentials.json:/etc/mcp-google-workspace/credentials.json:ro,Z \
  -v ./prompts:/etc/mcp-google-workspace/prompts:ro,Z \
  ghcr.io/fabiendupont/mcp-google-workspace:0.7.0 \
  --eager-tools --prompts-dir /etc/mcp-google-workspace/prompts

Building locally

podman build -t mcp-google-workspace:local .

The Dockerfile uses UBI 10 as the builder with system Rust and glibc-static for static linking.

Kubernetes

Manifests are in deploy/kubernetes/.

kubectl create namespace mcp-google-workspace
kubectl create secret generic mcp-gws-credentials \
  --from-file=credentials.json=./your-credentials.json \
  -n mcp-google-workspace
kubectl apply -k deploy/kubernetes/

What gets created

Probes

Multi-user

Each user gets their own Deployment, ConfigMap, and Secret. No shared state. See Architecture for the multi-user diagram.

Edit this page on GitHub