The gcloud Model Context Protocol (MCP) server enables AI assistants to easily interact with the Google Cloud environment using the gcloud CLI. With the gcloud MCP server you can:
- Interact with Google Cloud using natural language. Describe the outcome you want instead of memorizing complex command syntax, flags, and arguments.
- Automate and simplify complex workflows. Chain multiple cloud operations into a single, repeatable command to reduce manual effort and the chance of error.
- Lower the barrier to entry for cloud management. Empower team members who are less familiar with gcloud to perform powerful actions confidently and safely.
- Node.js: version 20 or higher
- gcloud CLI
To integrate this extension with Gemini CLI or Gemini Code Assist, run the setup command below. This will install the MCP server as a Gemini CLI extension for the current user, making it available for all your projects. Refer to Gemini CLI extension management for uninstall, disable, enable, and update instructions.
gemini extensions install https://github.com/gemini-cli-extensions/gcloudAfter the initialization process, you can verify that the gcloud-mcp server is configured correctly by running the following command:
gemini mcp list
> ✓ gcloud: npx -y @google-cloud/gcloud-mcp (stdio) - Connected
See github.com/googleapis/gcloud-mcp for instructions on using the gcloud MCP server with other AI clients.
| Tool | Description |
|---|---|
run_gcloud_command |
Executes a gcloud command. Some commands have been restricted from execution by the agent. See MCP Permissions for more information. |
The permissions of the gcloud MCP extension are directly tied to the permissions of the active gcloud account. To restrict permissions and operate with the principle of least privilege, you can authorize as a service account using impersonation and assign the service account a role with limited permissions.
By default, the gcloud MCP extension prevents execution of gcloud commands that don't make sense for AI agents. This is done to restrict commands that can run arbitrary inputs and initiate interactive sessions. See here for the list of denied commands.
We welcome contributions! Whether you're fixing bugs, sharing feedback, or improving documentation, your contributions are welcome. Please read our Contributing Guide to get started.
This repository is currently in preview and may see breaking changes. This repository is providing a solution, not an officially supported Google product. It may break when the MCP specification, other SDKs, or when other solutions and products change. See also our Security Policy.