Gemini Cloud Assist MCP

Gemini Cloud Assist MCP server

[!IMPORTANT]

Private Preview Notice

The Gemini Cloud Assist MCP server APIs are currently in Private Preview and are behind an allowlist. Please contact your Google Cloud account team to request access.

[!WARNING]

Deprecation Notice & Migration to Remote MCP Server

The Gemini Cloud Assist MCP server has migrated from a local Node.js architecture to a Remote MCP Server architecture. The older local Node.js server will lose support in the coming months.

To use the new Remote MCP Servers, please use version v0.8.0 or later. If you wish to continue using the legacy local server during the transition, please pin your configuration to older versions.

This server connects Model Context Protocol (MCP) clients to the Gemini Cloud Assist APIs. It allows you to use natural language to understand, manage, and troubleshoot your Google Cloud environment directly from the local command line.

The configurations in this repository connect to the Cloud Operator endpoint, https://geminicloudassist.googleapis.com/mcp/autocloud, which serves the Cloud Operator tools (ask_cloud_operator, search_operator_sessions, fetch_operator_session). The other Gemini Cloud Assist tools listed in the reference documentation stay on https://geminicloudassist.googleapis.com/mcp and are not reachable through these configurations.

[!NOTE]

The Google Cloud Platform Terms of Service (available at https://cloud.google.com/terms/) and the Data Processing and Security Terms (available at https://cloud.google.com/terms/data-processing-terms) do not apply to any component of the Gemini Cloud Assist MCP Server software.

To learn more about Gemini Cloud Assist, see the Gemini Cloud Assist overview in the Google Cloud documentation.

✨ Key features

Cloud Operator is an operator agent with live project context. Through ask_cloud_operator it handles:

• Troubleshoot issues: Incident and root-cause investigation in your Google Cloud environment.
• Review rollouts: Rollout readiness and monitoring review.
• Design and review infrastructure: Terraform best-practices review and generation, Kubernetes and Helm manifests, with results applied as reviewable file edits.
• Optimize costs: Analyze your spend and identify opportunities for efficiency.
• Get general assistance: Documentation-grounded how-to answers and guidance on Google Cloud best practices, architectures, and operations.

Quick start

Before you begin, ensure you have the following set up:

• Google Cloud SDK installed and configured.
• A Google Cloud project.
• The following IAM roles on your user account:
• roles/serviceusage.serviceUsageAdmin: Required to enable the Cloud Assist APIs.
• roles/geminicloudassist.user: Required to make requests to the Cloud Assist APIs.

Step 1: Authenticate to Google Cloud

The Gemini Cloud Assist MCP server uses local Application Default Credentials (ADC) to securely authenticate to Google Cloud. To set up ADC, run the following gcloud commands:

# Authenticate your user account to the gcloud CLI

gcloud auth login

Set up Application Default Credentials for the server.

gcloud auth application-default login

Configure your MCP client

The client-agent configuration depends on which agent you are using.

Cloud Operator plugin: Claude Code, ChatGPT Codex and Google Antigravity

[!IMPORTANT]

Private Preview. Cloud Operator and its MCP server are in Private Preview.

The Cloud Operator plugin (in plugins/cloud-operator; it installs as cloud-operator, so its commands are /cloud-operator:…) connects Claude Code, ChatGPT Codex or Google Antigravity to Cloud Operator (ask_cloud_operator). It:

• sets up the MCP connection to Cloud Operator, with a 30-minute tool timeout;
• asks which operator agent to use and checks it answers;
• saves your Google Cloud preferences and passes them to Cloud Operator;
• adds a skill that helps your agent decide when and how to use Cloud Operator.

To install it, see plugins/cloud-operator. In these agents, the plugin replaces the gemini-cloud-assist-mcp plugin at the root of this repository, so don't install both.

Antigravity

Add the following to your mcp_config.json:

"mcpServers": {

"gemini_cloud_assist": { "serverUrl": "https://geminicloudassist.googleapis.com/mcp/autocloud", "headers": {}, "authProviderType": "google_credentials" } }

Cursor

1. In your Google Cloud project, create an OAuth 2.0 client ID for a desktop app.
2. Configure URI://anysphere.cursor-mcp/oauth/callback as the redirect URL.
3. Add or merge the following configuration block:

{

"mcpServers": { "gemini_cloud_assist": { "url": "https://geminicloudassist.googleapis.com/mcp/autocloud", "auth": { "CLIENT_ID": "${env:OAUTH_CLIENT_ID}", "CLIENT_SECRET": "${env:OAUTH_CLIENT_SECRET}", "scopes": ["https://www.googleapis.com/auth/cloud-platform"] } } } }

Claude

Follow the configuration instructions for your specific application:

• Claude.ai
• Claude Code: see Cloud Operator plugin.

MCP Tools

Cloud Operator MCP tools

Cloud Operator is an agent accessible through a set of MCP tools. The agent invoked by MCP tool calls makes its own tool calls internally to Google Cloud. The /mcp/autocloud endpoint serves the following tools. They are in Private Preview and restricted to allowlisted projects.

The other Gemini Cloud Assist tools (ask_cloud_assist, design_infra, investigate_issue, invoke_operation, optimize_costs) are served by https://geminicloudassist.googleapis.com/mcp and are documented in the Gemini Cloud Assist MCP reference. They are not available through the configurations in this repository.

Note: These tools should not be treated as stable APIs. Parameters might be renamed or modified to account for the evolving capabilities of Gemini Cloud Assist.

Agent Skills

The Gemini Cloud Assist MCP tools leverage SKILL.md files to instruct your agent on how to properly use the tools. The skills help to guide your agent on when to delegate to Cloud Operator, passing relevant local information to it, and enabling explicit invocation.

Contributing

• If you encounter a bug, please file an issue on our GitHub Issues page.
• Before sending a pull request, please review our Contributing Guide.

License

This project is licensed under the Apache 2.0 License and provided as-is, without warranty or representation for any use or purpose. For details, see the LICENSE file.