# OpenCLI Agent Pack: Playwright MCP

Use this when an AI agent needs to work with `playwright-mcp`.

## What this CLI is for
The official CLI from Microsoft. Agent browser control, web testing, and browser automation from the terminal. Start with `npx @playwright/mcp@latest --help` and go from there. Runs entirely on your machine.

Best for: Giving agents browser automation through the Model Context Protocol.

## Agent readiness
Usable with guardrails (60/100)
- Supports non-interactive/scripted use.
- Works well in CI or repeatable automation.
- Can run locally with less credential exposure.

## Install
```sh
npm install -g @playwright/mcp
```

## Verify before real work
```sh
npx @playwright/mcp@latest --help
```
Expected signal: The Playwright MCP server help text prints successfully.

## Safe starting commands
```sh
npx @playwright/mcp@latest --help
```

## Guardrails for agents
- Prefer small commands and ask the agent to summarize plain text output.
- Network access is required; avoid leaking secrets in logs.
- Start read-only, then ask before mutations.

## Suggested agent instruction
You may use Playwright MCP (`playwright-mcp`) for giving agents browser automation through the model context protocol.. First install it if missing, then run the verify command. Start with read-only or inspection commands. Summarize what you found before changing anything. Ask for confirmation before commands that mutate remote state, spend money, deploy, delete data, merge code, or expose secrets.

Source: OpenCLI