Get Agent Loadout
Browser automationBeginnerDocs-checked, not yet hands-on tested

Browser automation agent (local)

Let an agent click through websites: test your app, fill forms, collect data.

Starter prompt

Open http://localhost:3000 and test the signup flow like a new user. Try one valid signup and three invalid ones. Report each step, what you expected, what happened, and take a screenshot of anything broken.

Config

.mcp.json

{
  "mcpServers": {
    "playwright": { "command": "npx", "args": ["@playwright/mcp@latest"] }
  }
}

Setup steps

  1. Add the Playwright MCP server.
  2. Start your app locally.
  3. Run the prompt; add --headless to the args to hide the browser.

Why this works

Accessibility snapshots give the model a structured view of the page, which is more reliable than screenshots alone.

Watch out

Don't let it log into accounts that hold real money or data without watching.

Related loadouts

Compare the gear: AI coding agents tier list · AI code review tier list · Browser automation for agents tier list