Google Agents CLI: A Meta-Tool That Turns Coding Agents into Agent Builders

Google unveiled agents-cli at Cloud Next.

agents-cli is a meta-tool that attaches Google Cloud AI agent design and deployment capabilities to coding agents like Gemini CLI, Claude Code, and Codex. Rather than having you write SDK code directly, it standardizes the process of deciding which components to combine, in what order, and with what configuration.

The core scope is divided into 7 skills.

• Workflow design
• ADK code writing
• Project scaffolding
• Evaluation: including LLM-as-judge
• Deployment: Agent Runtime, Cloud Run, GKE
• Gemini Enterprise publishing
• Observability: understanding state via logs and traces

Tool connectivity is also included.

• Supports MCP, A2A, and various connectors
• On the Google Cloud stack, it ties ADK, Agent Runtime, Agent Sandbox, and Gemini Enterprise together under a single command system
• Local development is possible with just an AI Studio API key
• A Google Cloud account is only required for cloud deployment
• Installation is a single line, uvx google-agents-cli setup, and requires Python 3.11+, uv, and Node.js
• For existing projects, scaffold enhance can retroactively add deployment configuration and CI/CD
• Can also run standalone in a terminal without a coding agent

It adopts a discovery-first design that surfaces the reasoning behind decisions, encouraging explanations not just of "what was done" but also "why that decision was made."

However, it is still at the Pre-GA stage and is distributed only as a pre-built .whl, not source code. Because it is strongly tailored to the Google Cloud stack, its applicability may be limited in multi-cloud or non-Google environments.