Enable coding assistants to build and deploy Google Cloud agents
You do not need to learn cloud configurations or framework syntax every time you create an agent. When you tell a coding assistant to "create an agent," it writes the necessary code and runs tests. This tool equips coding assistants with the knowledge required to develop, evaluate, and deploy agents on Google Cloud.
How it transfers expertise to assistants
This tool does not replace the coding assistant itself. Instead, it provides commands and skills that allow the assistant to use the Google Cloud Agent Development Kit (ADK) effectively. When a user requests an agent, the assistant follows the tool's instructions to set up the project structure and write the code. This enables the entire process, from development to evaluation, deployment, and monitoring, to be handled as a single flow. It removes the need for users to memorize the complex settings of individual cloud services.
Coverage of the entire agent development process
The tool provides specific features for each stage of agent development. Initially, it generates the project skeleton, and during development, it checks code quality. In the evaluation stage, it verifies agent responses against datasets and analyzes failure causes. During deployment, it uploads the agent to various Google Cloud environments and sets up CI/CD pipelines if needed. It also includes observability features to track and log agent behavior, making it easy to understand the agent's status after deployment.
Installation and usage environment
To get started, you need an environment with Python 3.11 or higher, uv, and Node.js installed. The installation process consists of two command steps. First, run uvx google-agents-cli setup to install the base tools, then register the skill with your coding assistant using npx skills add google/agents-cli. After that, you can immediately use it by opening your preferred coding assistant, such as Antigravity CLI, Claude Code, or Codex. If you only want to develop locally, you only need an AI Studio API key, but using the cloud deployment feature requires a Google Cloud project.
Data collection and privacy
By default, the tool records usage signals when commands are executed. This record includes only the command name, exit code, error type, duration, CLI and Python versions, and operating system information; it never collects file contents or input arguments. This data is stored in Cloud Logging within the user's Google Cloud project, where users can query it directly. If you do not want this, you can completely disable collection by setting the AGENTS_CLI_TELEMETRY=0 or DO_NOT_TRACK=1 environment variables. You can check the current collection status using the agents-cli info command.
Google Cloud dependencies and cost responsibility
Because this tool operates based on Google Cloud APIs, resources created during deployment belong to the user's Google Cloud project. Therefore, users are directly responsible for the costs associated with deployed agents and related infrastructure. No Google Cloud costs are incurred if you only perform development and evaluation locally, but you must review the Google Cloud service terms when using actual deployment and cloud features. It is also advisable to clearly understand that this tool complements the capabilities of existing assistants rather than replacing them.
