Browser Automation

browser(action=...) provides DOM-aware browser control via Chrome DevTools Protocol (CDP).

Overview

The Browser tool is the middle-layer interaction tool — it understands web page DOM structure and can precisely operate elements via CSS selectors or Accessibility Tree refs. Suitable for all web interaction scenarios.

Available Actions

Action	Description	Key Params
`navigate`	Go to URL	`url`, `new_tab`
`screenshot`	Page screenshot	`url`/`html`/`file_path`, `full_page`
`snapshot`	Get Accessibility Tree	`selector`
`click`	Click element	`ref`, `button`
`type`	Type keystroke-by-keystroke	`ref`, `text`, `clear`, `submit`
`fill`	Set value instantly	`ref`, `text`
`scroll`	Scroll	`direction`, `amount`, `ref`
`press_key`	Press key	`key`
`tabs`	Tab management	`tab_action`, `tab_index`
`set_html`	Inject HTML via CDP	`html`
`go_back`	Navigate back	—

Screenshot: Three Modes

1. URL Mode (running page)

browser(action="screenshot", url="http://localhost:5173")

2. HTML Injection Mode (zero-server)

browser(action="screenshot", html="<html><body><h1>Hello</h1></body></html>")

Injects HTML via CDP Page.setDocumentContent. No server needed.

3. File Path Mode (local file)

browser(action="screenshot", file_path="/home/user/workspace/demo.html")

Auto-starts bun serve and captures the file.

Accessibility Tree Workflow

# 1. Get page structure
browser(action="snapshot")
# → [0] button "Submit" [1] input "Email" [2] link "Home" ...

# 2. Operate by ref
browser(action="fill", ref="input[1]", text="user@example.com")
browser(action="click", ref="button[0]")

Execution Backends

Environment	Backend	Notes
Cloud Sandbox	CDP on sandbox Chromium	Direct CDP control
Desktop	Electron in-app browser	WebSocket MCP

Difference from `chrome`

Aspect	browser	chrome
Environment	Sandbox/controlled browser	User’s real Chrome
Auth state	Clean (no logins)	User’s logged-in sessions
Confirmation	Not required	4-level policy
Tab management	AI creates freely	Discover → Claim
Risk	None (isolated)	Real-world actions

Computer Use Chrome Extension

Get Started

Core Concepts

Chat with Experts

Creation & Automation

Plugin Tools

Memory & Personalization

Marketplace & Content

IM Channels

Account & Billing

Team

How It Works

Help

Browser Automation

Browser Automation

Overview

Available Actions

Screenshot: Three Modes

1. URL Mode (running page)

2. HTML Injection Mode (zero-server)

3. File Path Mode (local file)

Accessibility Tree Workflow

Execution Backends

Difference from `chrome`

​Browser Automation

​Overview

​Available Actions

​Screenshot: Three Modes

​1. URL Mode (running page)

​2. HTML Injection Mode (zero-server)

​3. File Path Mode (local file)

​Accessibility Tree Workflow

​Execution Backends

​Difference from chrome

Browser Automation

Overview

Available Actions

Screenshot: Three Modes

1. URL Mode (running page)

2. HTML Injection Mode (zero-server)

3. File Path Mode (local file)

Accessibility Tree Workflow

Execution Backends

Difference from `chrome`