Getting Started with Kiro CLI

After work you need an agent that lives in the project directory and talks natural language in the terminal. Kiro CLI is AWS’s agentic CLI: interactive chat, a rich TUI, headless mode for scripts, MCP connections, and — when you want an editor bridge — Agent Client Protocol (ACP) so JetBrains, Zed, and other ACP clients can spawn kiro-cli acp.

This guide is the evening install + one small Python task. It is not a full AWS DevOps playbook; MCP and custom agents can wait for night two.

What Kiro CLI is

Official docs describe Kiro CLI as AI-assisted development in your terminal: build, test, and automate with natural language. Shared capabilities with other Kiro surfaces include steering files, hooks, MCP, custom agents, skills, sub-agents, models, and permissions. CLI-unique extras include interactive chat, terminal UI, headless/KIRO_API_KEY flows, voice mode, and ACP.

ACP matters if you already live in JetBrains or Zed: run kiro-cli acp (or kiro-cli acp --agent my-agent) and point the editor at that command over stdio JSON-RPC. Night one can stay in the plain TUI with kiro-cli inside a practice folder.

Who it is for

Learners who can open a terminal, install via the official script, and sign in with AWS Builder ID or another supported IdP (Google/GitHub/org flows per authentication docs). You want agent help on local .py files after dinner — not a multi-account infra migration tonight.

Start steps (one evening)

1. Install. macOS/Linux/Windows (per CLI docs): curl -fsSL https://cli.kiro.dev/install | bash. Confirm with kiro-cli --version (or kiro-cli --help).
2. Log in. kiro-cli login and complete the browser (or device-code) flow. Remote SSH cannot open a local browser; follow the URL/code path in the auth docs. Interactive evening sessions use login; reserve KIRO_API_KEY for headless/CI later.
3. Enter a practice project. cd /path/to/tiny-python-lab then kiro-cli.
4. Orient before editing. Ask what the folder does or which file is the entry point. Let it read.
5. One bounded edit. Ask for a small, testable change; review tool calls; run python yourself.

Optional later the same week: wire ACP in Zed/JetBrains, or add MCP (including AWS MCP patterns from AWS docs) — not required for the first win.

First tiny exercise (15–20 minutes)

Create numbers.py:

def mean(values):
    return sum(values) / len(values)

if __name__ == "__main__":
    print(mean([2, 4, 6]))

Run kiro-cli in that folder. Ask: Add a median for a non-empty list of numbers, call it from main after mean, keep under 40 lines, explain before editing.

Verify:

python numbers.py

Exit when mean and median print correctly. Do not open a second agent or install five MCP servers before you sleep.

Common pitfalls

• Skipping login. Install alone does not authenticate; run kiro-cli login before blaming models.
• Production AWS credentials on night one. You do not need a full Agent Toolkit / MCP AWS stack to edit a practice script. Add cloud tools when you have a clear, small AWS task.
• Confusing IDE vs CLI. The kiro launcher may target IDE by default depending on integrations; for this path use kiro-cli explicitly.
• Vague prompts. “Improve everything” at 11pm creates unreadable diffs.
• Diving into ACP JSON first. Learn the TUI loop before spawning agents from editor config files.

Next: Python under the agent

Kiro speeds editing and exploration. You still judge every change. Introduction to Python Programming fits after-work blocks. Check the course page for the current enrollment window.

Tonight: install, login, one explainer, one small edit, one python success. Stop there.