agents-cli Review (2026): Pricing, Features & Honest Verdict

TLDR

agents-cli is Google’s free, open-source (Apache-2.0) command-line tool that turns coding agents like Claude Code into experts at scaffolding, evaluating, and deploying AI agents to Google Cloud. It handles the plumbing, evals, and CI/CD so you ship to production faster. Best for: teams building ADK agents on GCP. Price: From $0 (free plan: yes). Rating: 7.9/10

What is agents-cli?

agents-cli is a command-line tool plus a skills package that automates the full lifecycle of building AI agents on Google Cloud, from scaffolding a project to running evaluations to deploying to production. The pitch is simple: you can build a demo agent in an afternoon, but getting it to production is weeks of plumbing, SDKs, MCP servers, auth, telemetry, and CI/CD. agents-cli collapses that into a handful of commands, and it is explicitly designed to be driven by your coding agent rather than typed by hand.

Google’s Developer Relations team launched it publicly in April 2026 and it hit Product Hunt in July 2026 with the tagline “the CLI your coding agent uses to ship agents.” It sits on top of Google’s Agent Development Kit (ADK) and installs the right CLAUDE.md or GEMINI.md skill files so tools like Claude Code, Antigravity, and Codex know exactly how to use it. It runs headless, so your coding agent can scaffold, test, and iterate on its own. The repo has crossed 5,000 GitHub stars, which is fast traction for a niche developer tool.

What Are agents-cli’s Key Features?

One-command scaffolding

Running agents-cli create generates a production-ready agent project: the agent code, tools, tests, a Dockerfile, observability wiring, Terraform, and CI/CD. You can start from expert-crafted templates, a GitHub repo, or adk-samples shortcuts like adk@data-science. Instead of stitching together ten docs tabs, you get a working skeleton that already follows Google’s recommended structure.

Built-in evaluation harness

The eval command is the standout. It runs your agent over a dataset (generate), grades traces against metrics like final_response_quality and grounding (grade), diffs two runs (compare), and can even synthesize multi-turn test conversations and optimize prompts using the GEPA framework. This turns “does it feel better?” into actual numbers you can iterate against, which most agent tooling skips entirely.

Deploy to Agent Runtime, Cloud Run, or GKE

One agents-cli deploy ships your agent to Vertex AI Agent Engine (Agent Runtime), Cloud Run, or GKE, then publish gemini-enterprise registers it with Gemini Enterprise. The infra command provisions the underlying Google Cloud resources and sets up CI/CD via Cloud Build or GitHub Actions, complete with staging and prod environments and Terraform state management.

Agent Mode and Human Mode

You can run everything interactively with guided prompts (Human Mode, via -i), or fully headless with all flags supplied and –auto-approve to skip confirmations. That headless path is what lets a coding agent self-optimize against success criteria you define without stopping to ask questions.

How Much Does agents-cli Cost?

agents-cli is completely free and open-source under the Apache-2.0 license. You install it with a single line, uvx google-agents-cli setup, and there is no license fee, subscription, or seat cost for the tool itself. Running agents locally works with a free AI Studio API key. The only money involved is your own Google Cloud usage, so if you deploy to Cloud Run, GKE, or Vertex AI Agent Engine you pay Google Cloud’s standard metered rates for whatever compute and inference you consume. Compare that to a commercial agent platform like LangSmith, whose paid plans start around $39 per user per month, and agents-cli’s tool cost of $0 is hard to beat, though you are trading dollars for a hard dependency on the Google Cloud ecosystem.

Who Is agents-cli Best For?

agents-cli is aimed squarely at developers and teams who are committed to Google Cloud and want to ship ADK-based agents to production without hand-rolling the infrastructure. It rewards people already using a coding agent in their workflow.

• Use agents-cli if you are building agents on Google Cloud, you use the Agent Development Kit, and you want evals and CI/CD baked in from day one, especially if a coding agent is doing the heavy lifting.
• Skip agents-cli if you are not on Google Cloud, you build on OpenAI or Anthropic stacks directly, or you want a cloud-agnostic framework. The tool is deliberately GCP-first and offers little value outside that ecosystem.

Best agents-cli Alternatives

agents-cli is Google-Cloud-specific, so the right alternative depends on where you deploy and how much observability you need.

LangChain / LangGraph

The most popular open-source framework for building agent workflows, cloud-agnostic and free to self-host. It gives you more flexibility across providers but leaves deployment, evals, and infra largely up to you, whereas agents-cli hands you those on GCP out of the box.

LangSmith

LangChain’s commercial observability and evaluation platform, with a free developer tier and paid plans from around $39 per user per month. It is stronger on tracing and eval dashboards across any stack, but it is a paid SaaS layer rather than a free deploy-anywhere CLI.

Agent Starter Pack

Google’s own template collection (also free and open-source) for scaffolding production agents on Google Cloud. It overlaps heavily with agents-cli’s create command, but agents-cli wraps scaffolding, evals, and deployment into one lifecycle tool rather than just templates.

Final Verdict: Is agents-cli Worth It?

If you are shipping agents on Google Cloud, agents-cli is close to a no-brainer: it is free, Apache-licensed, actively maintained by Google, and it removes the most tedious part of agent development, the production plumbing. The built-in eval harness alone, with grading, comparison, and GEPA-based prompt optimization, is more rigorous than what most teams build themselves, and the headless mode makes it a genuinely good fit for a coding-agent-driven workflow.

The catch is the tight coupling to Google Cloud and ADK. This is not a neutral, portable tool, it is a very good on-ramp into Google’s agent ecosystem, and that lock-in is the price of the convenience. If you are already on GCP, install it today. If you are cloud-agnostic or invested in OpenAI or Anthropic tooling, this one is not for you, and a framework like LangGraph will serve you better.