agents-cli
The CLI your coding agent uses to ship AI agents on Google Cloud
Overview
agents-cli exists to close the gap between building a demo AI agent, which takes an afternoon, and getting one into production, which usually means weeks of plumbing: SDKs, MCP servers, auth, telemetry, CI/CD, and documentation. It installs with a single `uvx google-agents-cli setup` command and then works through your existing coding assistant, so you can ask something like "build an SRE agent that reads logs and drafts an incident report" and have it scaffold the whole project, run evaluations against a dataset, and deploy to Vertex AI Agent Engine, Cloud Run, or GKE.
It's free and open source under Apache-2.0, and there's a local option using an AI Studio API key so you can develop without a Google Cloud project at all. The moment you deploy for real, though, you're paying standard Google Cloud charges for whatever compute and services your agent uses, so the tool itself costs nothing but the infrastructure it provisions doesn't. This is squarely a developer tool tied to Google's cloud ecosystem, if your team is on AWS or Azure, most of the deployment tooling here won't apply directly.
Key Features
Project Scaffolding: Generates a full agent project including tools, tests, Dockerfile, observability, and Terraform in one command.
Built-In Evaluation: Runs the agent over a dataset and scores it against success criteria you define.
One-Command Deployment: Ships to Vertex AI Agent Engine, Cloud Run, or GKE without manually wiring up infrastructure.
Coding Assistant Native: Designed to be driven by Claude Code, Codex, or Antigravity CLI, so your existing agent can self-optimize based on eval results.
Headless Operation: Runs without a UI, so it can be chained into automated workflows your coding agent already manages.
Pricing
Starting price
Free and open source (Apache-2.0); you pay standard Google Cloud charges for any resources you actually deploy
agents-cli: Free and open source under Apache-2.0. Local development possible at no cost using a free AI Studio API key.
Production Deployment: No separate license fee, but you pay standard Google Cloud charges for Vertex AI Agent Engine, Cloud Run, or GKE resources your agent actually uses.
Disclaimer: pricing may change, confirm current Google Cloud service rates before deploying anything to production.
Pros
Free And Open Source: No license cost, Apache-2.0, and free local development via an AI Studio API key.
Cuts Real Setup Time: Scaffolds the infrastructure plumbing that normally eats weeks of an agent project.
Built-In Evaluation: Scoring agents against a dataset before deployment is a genuinely useful step most quick-start tools skip.
Coding Assistant Integration: Works with tools developers already use daily instead of requiring a separate interface.
Cons
Google Cloud Lock-In: Deployment targets are Vertex AI, Cloud Run, and GKE only, no support for other clouds.
Developer-Only: Requires Python, uv, and Node.js set up locally, not approachable for non-engineers.
Cloud Costs Are Separate: The free tool can still lead to real Google Cloud bills once you deploy anything at scale.
Very New: Recently launched, so best practices, community support, and long-term maintenance are still unproven.
What Makes It Unique
Built To Be Driven By Your Coding Agent: Runs headless so tools like Claude Code or Codex can scaffold, evaluate, and redeploy an agent on their own based on eval scores, rather than you clicking through a console.
Kay Score
7.6
/ 10
Tool Information
Pricing
Free and open source (Apache-2.0); you pay standard Google Cloud charges for any resources you actually deploy
Category
Automation & AI Agents
Platform
Web / iOS / Android
Last Updated
