browser

The fastest, most token-efficient way for AI agents to control Chrome.

Benchmarks in progress: Early testing shows 3x faster than agent-browser for a variety of usecases, but does not yet support as wide of a feature set.

Experimental: Commands are stable, but their output format is subject to change as we optimize for AI agent effectiveness.

Install

curl -fsSL https://raw.githubusercontent.com/camhahu/browser/main/packages/cli/install.sh | bash

Point your agent to the skill file to get started, or install it directly:

browser add-skill opencode            # Add to current project
browser add-skill --global opencode   # Add globally

browser add-skill amp        # Amp
browser add-skill codex      # OpenAI Codex
browser add-skill claude     # Claude Code
browser add-skill cursor     # Cursor
browser add-skill goose      # Goose
browser add-skill github     # GitHub Copilot
browser add-skill vscode     # VS Code

Update

browser update

Example

browser open https://example.com                      # Opens page, shows outline
browser click "Products" && browser text ".product-list"
browser stop

Commands

Browser Control

Command	Description
start [--headed|--headless]	Start browser (default: headed)
stop	Stop the browser

Navigation

Command	Description
open <url>	Open URL in new tab (auto-starts headless browser)
navigate <url>	Navigate active tab to URL
back	Go back in history
forward	Go forward in history
refresh	Reload active tab

Tab Management

Command	Description
tabs	List all open tabs
active	Print active tab ID
use <tab-id>	Switch to specific tab
close [tab-id]	Close tab (default: active)

Interaction

Command	Description
click <selector>	Click element (label, text, or CSS)
type <text> [selector]	Type text or send keys (e.g. Enter, ctrl+c)
hover <selector>	Move mouse to element
drag <source> <target>	Drag element to another element
select <selector> <value>	Select dropdown option
scroll <target>	Scroll to top, bottom, or CSS selector
wait <selector>	Wait for element (15s timeout)

Content Extraction

Command	Description
outline [-a [depth]]	Show interactive elements (or all with -a)
text [-l chars] [selector]	Get text content (default: 2000 chars)
html [-l chars] [selector]	Get HTML content
find <selector>	Find elements matching selector
url	Print active tab URL
title	Print active tab title

Screenshots

Command	Description
screenshot [-f format] [-o path] [name]	Capture screenshot (png/jpeg/webp)

Debugging

Command	Description
console [-l count] [-t types]	Show console output
network [-f pattern] [-t types] [id]	List/inspect network requests
eval <js>	Evaluate JavaScript

State & Storage

Command	Description
cookies [name]	List cookies or get specific value
storage [key]	List localStorage or get specific value

Viewport & User Agent

Command	Description
viewport desktop|tablet|mobile	Set viewport preset
viewport set <width> <height>	Set custom viewport
useragent macos|windows|iphone|android	Set user agent preset
useragent custom <string>	Set custom user agent

Configuration

Command	Description
config	Show all config
config set <key> <value>	Set config value
config unset <key>	Remove config value
config clear-profile	Clear browser profile (cookies, logins)

Utilities

Command	Description
add-skill <target>	Install skill file for AI agent
update	Update to latest version

Selectors

browser click l3              # Label from outline (fastest)
browser click "Sign up"       # Text match (exact first, then partial)
browser click ".btn-primary"  # CSS selector

Skill File

See packages/cli/skill/SKILL.md for the AI agent skill file with usage patterns.

Contributing

See docs/TODO.md for planned features and docs/BUGS.md for known issues.

bun install
bun run build
bun run test

The build outputs to ./dist/browser. You can symlink it for local testing:

sudo ln -s $(pwd)/dist/browser /usr/local/bin/browser

License

MIT