Agents CLI is a CLI and a set of skills for building, evaluating, and deploying
agents on Google Cloud, with [ADK](https://adk.dev/), LangChain or the Claude
Agent SDK. Agents CLI handles everything around your agent: scaffolding,
evaluation, deployment, and observability.
## What your coding agent does
1. **Installs Agents CLI**
Reads [setup.md](../../setup.md) and runs `uvx google-agents-cli setup`,
which installs the `agents-cli` command and its skills.
1. **Asks for your language**
Python or Go, with ADK as the framework. For LangChain or the Claude Agent
SDK, it scaffolds from a first-party template.
1. **Designs the agent with you**
Asks what the agent should do, which tools it calls and how to judge
success, then scaffolds the project and writes the agent with an eval
dataset.
1. **Connects a model**
A Gemini API key from AI Studio, or a Google Cloud project. See
[Authentication](https://google.github.io/agents-cli/guide/authentication/index.md). The API key works without code
changes in ADK Python and LangChain projects. An ADK Go project also needs
a change to `app/agent.go` ([details](https://google.github.io/agents-cli/guide/project-structure/index.md#env)). The
Claude template needs a Google Cloud project.
1. **Opens the playground**
The `agents-cli playground` command starts the local playground and
prints its URL (`http://127.0.0.1:8080/...` by default) so you can talk to
your agent. ADK templates serve the ADK web UI and the LangChain template
runs the LangGraph dev server. The Claude template has no playground
([First-party frameworks](https://google.github.io/agents-cli/guide/extensions/first-party/index.md)).
## Evaluate and deploy
Keep going in the same session:
```text
Write evals for this agent and run them
```
```text
Deploy this to Cloud Run
```
See [Evaluation](https://google.github.io/agents-cli/guide/evaluation/index.md) and [Deployment](https://google.github.io/agents-cli/guide/deployment/index.md) for what
happens under the hood.
## What the skills teach your coding agent
Agents CLI installs 7 skills:
| Skill | What your coding agent learns |
|-------|-------------------------------|
| `google-agents-cli-workflow` | Development lifecycle, code preservation, model selection |
| `google-agents-cli-adk-code` | ADK Python and Go APIs: agents, tools, orchestration, callbacks |
| `google-agents-cli-scaffold` | Project scaffolding: `create`, `enhance`, `upgrade` |
| `google-agents-cli-eval` | Evaluation lifecycle: datasets, metrics, generate and grade, compare, analyze, optimize |
| `google-agents-cli-deploy` | Deployment: Agent Runtime in Gemini platform, Cloud Run, GKE, CI/CD |
| `google-agents-cli-publish` | Registration in the Gemini app |
| `google-agents-cli-observability` | Cloud Trace, logging, third-party integrations |
See the [skills reference](https://google.github.io/agents-cli/reference/skills/index.md) for details.
## Run the steps yourself
### Prerequisites
**Required:** [uv](https://docs.astral.sh/uv/getting-started/installation/). It downloads Python 3.11 or later if you don't have it.
**Recommended:** [Node.js](https://nodejs.org/en/download), so setup installs the skills with `npx skills` and `agents-cli update` can update them. Without it, setup copies the skills instead.
**Optional (for deployment):** [Google Cloud SDK](https://cloud.google.com/sdk/docs/install), [Terraform](https://developer.hashicorp.com/terraform/downloads)
### Install
```bash
uvx google-agents-cli setup
```
The setup command installs the CLI and the skills for the coding agents it finds on your
machine.
??? info "Alternative installation methods"
**pipx:** `pipx install google-agents-cli && npx skills add google/agents-cli`
**venv + pip:** `pip install google-agents-cli && npx skills add google/agents-cli`
**Skills only:** `npx skills add google/agents-cli`
**Platform support:** macOS, Linux, and Windows (WSL 2). Native Windows is not officially supported.
### Authenticate
If you're already authenticated with `gcloud`, Agents CLI picks up your
Application Default Credentials automatically.
Otherwise, the quickest option is a Gemini API key from [AI Studio](https://aistudio.google.com/apikey):
```bash
export GEMINI_API_KEY="your-key-here"
```
In a scaffolded Python project, also comment out the `GOOGLE_*` lines in `.env` so the agent calls the Gemini API instead of Google Cloud. A Go project also needs a change to `app/agent.go`; see [`.env`](https://google.github.io/agents-cli/guide/project-structure/index.md#env). See [Authentication](https://google.github.io/agents-cli/guide/authentication/index.md) for full details.
### Build with your coding agent
Open your coding agent and check that it lists the `google-agents-cli-*` skills.
In Claude Code, run `/skills`. Most other coding agents have a similar command
or a settings panel. Then ask it to build something:
```
Build a support agent that answers questions from our docs
```
### Build without a coding agent
You can drive the entire workflow from your terminal, with no coding agent.
```bash
# Create a minimal agent project
agents-cli create my-agent --prototype --yes
# Install dependencies and start the dev playground
cd my-agent
agents-cli install
agents-cli playground
```
In ADK templates, the playground command starts the ADK web playground, with hot reload in Python projects. Other templates differ; see [First-party frameworks](https://google.github.io/agents-cli/guide/extensions/first-party/index.md). Open the URL the command prints (`http://127.0.0.1:8080/...` by default).
For a full walkthrough, see the [manual workflow tutorial](https://google.github.io/agents-cli/guide/hands-on-tutorial/index.md).
## Templates and status
Status as of October 2026.
| Language | Template | Status | Notes |
|----------|----------|--------|-------|
| Python | `adk_python` (alias `adk`) | Default | ADK; A2A built in |
| Go | `adk_go` | Supported | ADK Go; API key needs a code change |
| Python | First-party template path | See [First-party frameworks](https://google.github.io/agents-cli/guide/extensions/first-party/index.md) | LangGraph, Claude Agent SDK |
| Your choice | Your own template | Supported | `--agent github.com/org/repo/path@tag`; see [Authoring](https://google.github.io/agents-cli/guide/extensions/authoring/index.md) |
Pass the template with `--agent`. Without it, `agents-cli create` uses `adk_python`.
## Demo
## Next steps
- [Tutorial: Build your first agent](https://google.github.io/agents-cli/guide/quickstart-tutorial/index.md): build, evaluate, and deploy with your coding agent
- [Tutorial: Manual workflow](https://google.github.io/agents-cli/guide/hands-on-tutorial/index.md): type every command yourself
- [Use cases](https://google.github.io/agents-cli/guide/use-cases/index.md): agent patterns people build
- [Project structure](https://google.github.io/agents-cli/guide/project-structure/index.md): what each generated file does
- [Extend the starter agent](https://google.github.io/agents-cli/guide/development/index.md#extend-the-starter-agent): the recipes your coding agent studies to add capabilities
- [Development guide](https://google.github.io/agents-cli/guide/development/index.md): the full development workflow
- [CLI reference](https://google.github.io/agents-cli/cli/index.md): all commands and flags
- [Security and trust](https://google.github.io/agents-cli/guide/security/index.md): what setup installs and runs, and how to limit it
!!! note "Share what you build"
Built something interesting with Agents CLI? Share your project at [agents-cli@google.com](mailto:agents-cli@google.com).