Google Cloud MCP Servers: Tools, Setup and Security

Google Cloud MCP servers let AI apps reach Google Cloud products such as BigQuery, Cloud Run, Compute Engine and Cloud Logging through remote endpoints that Google runs. Google also publishes local, open-source servers, including one that runs gcloud CLI commands.

Checked against: Google Cloud MCP servers overview on 28 October 2026. eSEOspace did not install or test this server.

At a glance

What it can do

Remote servers

Google's supported products page lists dozens of endpoints. Examples for infrastructure teams:

Some, including Cloud Billing, API keys and Cloud Asset Inventory, are labeled Preview. Each product links to its own MCP reference with the tool list. Google notes that some servers offer toolsets, and "each toolset has its own HTTP endpoint and functions as a virtual MCP server."

Local gcloud servers

The gcloud-mcp README lists four local servers:

• gcloud: one tool, run_gcloud_command, which runs gcloud commands. Some commands are blocked by default.
• observability: tools such as list_log_entries, list_time_series, list_alert_policies and get_trace.
• storage: tools such as list_objects, write_object, delete_object, create_bucket and delete_bucket.
• backupdr: backup vault and plan tools, including restore_backup and delete_backup.

How to connect it

For remote servers, Google's enable guide says to enable the product API in your project. You need the Service Usage Admin role:

gcloud services enable SERVICE_NAME

For BigQuery, the service name is bigquery.googleapis.com. Google then points you to separate pages to authenticate and configure your AI app. Servers can also be managed in Agent Registry.

For the local gcloud server, you need Node.js 20 or higher and the gcloud CLI. The README gives this config for most clients:

"gcloud": {
  "command": "npx",
  "args": ["-y", "@google-cloud/gcloud-mcp"]
}

For Gemini CLI, Google documents a one-line installer:

npx @google-cloud/gcloud-mcp init --agent=gemini-cli

Security notes

• IAM controls access. Google says "IAM allow and deny policies help you secure Google Cloud MCP servers." Use deny policies to block risky tools for agent identities.
• Model Armor. Google offers Model Armor to "scan MCP calls and responses," which helps against prompt injection through tool results.
• Audit logging. Google links to a dedicated audit logging page for its MCP servers.
• Local server permissions. The gcloud server inherits the active gcloud account's permissions. Google recommends impersonating a service account with a limited role.
• Destructive local tools. The storage and backupdr servers include delete and restore tools. Google's README marks several of them in bold as deleting data.
• Support status. The gcloud-mcp repository "is currently in preview and may see breaking changes." Google says it is not covered by the Google Cloud Terms of Service.

Read excessive permissions in MCP servers and the MCP security hub before you grant an agent project-wide roles.

Alternatives

• AWS MCP Server for AWS accounts.
• Azure MCP Server for Azure subscriptions.
• Kubernetes MCP Server (Community) if you need cluster-level access beyond GKE's own server.

Next step

Pick one product endpoint, enable it in a test project and give the agent a narrow role. Our MCP readiness audit helps you choose that first project. See the server directory for related profiles.

Frequently asked questions