Agent Browser Control
Once the Verne MCP server is enabled for an agent, that agent gains a set of browser tools. It operates on the same tabs open in your embedded browser, so you can watch every action.
To enable MCP for an agent, go to Settings › Agents. See Hooks and MCP setup.
Browser tools
| Tool | What it does |
|---|---|
browser_list |
List all open browser tabs with their IDs and URLs |
browser_open |
Open a URL in a new tab |
browser_navigate |
Navigate an existing tab to a URL |
browser_back |
Go back in a tab’s history |
browser_forward |
Go forward in a tab’s history |
browser_reload |
Reload a tab |
browser_snapshot |
Capture the accessibility tree for a tab, with element refs the agent can act on |
browser_click |
Click an element by its ref from a snapshot |
browser_fill |
Fill a text input with a value |
browser_select |
Choose an option in a dropdown |
browser_screenshot |
Take a JPEG screenshot of a tab |
browser_network |
Retrieve recent network requests made by a tab |
browser_console |
Retrieve console messages from a tab |
browser_wait |
Wait until a tab reaches a given state: load, networkidle, or a CSS selector |
What agents can do with this
An agent that just changed UI code can open the app in the browser, reload, take a screenshot, and confirm the change looks right. It can reproduce a bug by navigating to a specific page, filling in a form, and capturing what the console logs. It can also read a rendered page through the accessibility snapshot when scraping raw HTML would miss dynamic content.
Because the agent acts on the tabs you already have open, you see exactly what it sees. You can step in at any point.
Next steps
- Hooks and MCP setup: enable MCP for an agent.
- MCP server reference: full reference for all Verne MCP tools.
- Embedded browser: how the browser panel works.