Google Cloud Introduces CLI Remote MCP Server in Preview

Google Cloud has introduced the Google Cloud CLI remote MCP server in public preview, expanding its ecosystem of managed remote MCP servers. This new service packages hundreds of gcloud and bq (BigQuery) command-line tools into a single Model Context Protocol (MCP) server, allowing AI agents to manage Google Cloud infrastructure and execute advanced BigQuery workflows securely.

Why CLI for AI Agents

The server addresses the challenge of standardizing how agents interact with backend systems by leveraging two key advantages of CLI operations:

• Higher-level abstractions: CLI commands bundle complex multi-step workflows and validation checks into unified commands, avoiding the need for multi-step API orchestration.
• Leverages model training: LLMs are heavily pre-trained on public command-line documentation and syntax, making CLI invocation intuitive and highly accurate for models.

Security and Governance

To ensure enterprise-grade security, the server isolates execution in a network-restricted proxy boundary on Google Cloud infrastructure, eliminating the need for local CLI binaries in agent environments. Key security features include:

• Zero ambient credentials: Authentication and authorization are handled through Agent Identity, OAuth 2.0, and Identity and Access Management (IAM).
• Strict policy enforcement: Commands run with the permissions of the authenticated caller, enforcing both standard IAM permissions and organization policy constraints.
• Model Armor integration: The server integrates with Model Armor to screen LLM prompts and responses, protecting against risks like prompt injection.
• Cloud audit logging: Every tool invocation is logged to Audit Logs, providing visibility into caller identities and IAM decisions without exposing sensitive payloads.

Tools and Availability

The server exposes two primary tools: run_gcloud_command for broad infrastructure management and run_bq_command for advanced BigQuery tasks such as resource allocation, job monitoring, and task scheduling. The service is available today in public preview at https://cloudcli.googleapis.com/mcp. There is no additional charge for the MCP server itself; users pay only for the GCP resources created and applicable data transfer costs.