Drive the Browser with an Agent

Verne’s embedded browser shares a tab with your agent. The agent can navigate, click, fill forms, and take screenshots while you watch.

1. Enable MCP for your agent

The browser tools are exposed over MCP. Open Settings › Agents, find your agent, and enable the Verne MCP server for it. See Hooks & MCP Setup for the full configuration steps, or the MCP server reference for the list of available tools.

2. Open your app in the browser

Open the embedded browser from the panel on the right. Navigate to your app’s local URL, or let the agent open it as part of the task.

3. Ask the agent to interact

Give your agent a prompt that describes what to verify. The agent uses its browser tools to navigate pages, click elements, fill in fields, and capture screenshots. The agent drives the same tab you’re watching, so every action is visible.

$ claude

Example prompt: “Open http://localhost:3000, click the Sign In button, fill in the test credentials, and confirm the dashboard loads.”

4. Review the result

The agent reports back what it found. Screenshots appear in the terminal output. If something is wrong, ask the agent to fix it and re-verify; it can loop through the browser steps without you switching context.

See Agent Browser Control for the full set of tools the agent can use.

Next steps