oh-my-browser

中文文档

Let your AI Agent use your real browser to search and interact with the web -- with your logins, cookies, and locale intact.

Features

• Authenticated Web Access -- Agent uses your browser session to access login-required pages (Gmail, Notion, internal tools, etc.)
• Multi-Engine Search -- Search via Google, Baidu, Sogou, or your browser's session for personalized results
• Page Reading -- Extract clean content from any webpage, with format options (markdown, text, HTML)
• Browser Automation -- Click, fill, scroll, screenshot, run JS -- full browser interaction via your real Chrome
• Element Discovery -- Find interactive elements by label, type, or region before interacting
• Accessibility Snapshots -- Stable element refs for SPA/dynamic pages where CSS selectors break
• Zero Config -- No API keys to configure. Install, authorize, and go.

How It Works

Agent --> oh-my-browser plugin --> omb CLI (subprocess) --> API --> your Chrome extension

The plugin spawns an omb CLI subprocess and communicates via MCP (stdio JSON-RPC). All browser operations execute through the Oh My Browser Chrome extension in your real browser.

Installation

One-line install (recommended)

# macOS / Linux
curl -fsSL https://api.omb.org.cn/install | bash

# Windows (PowerShell)
irm https://api.omb.org.cn/install | iex

The install script will interactively guide you through:

1. Download omb CLI -- the command-line tool
2. Login -- opens your browser for authorization
3. Install OpenClaw plugin -- auto-detects OpenClaw instances
4. Download Chrome extension -- the bridge between Agent and your browser

Or install plugin only

openclaw plugins install oh-my-browser

If omb CLI is not yet installed, run the one-line install above to set it up.

Login

If you skipped login during installation, or your session expired:

omb login

This opens your browser for authorization. In an OpenClaw Agent session, the Agent will guide you through authorization automatically.

Then restart the gateway:

openclaw gateway restart

Tools

Tool	Description
omb_search	Search the web (auto, browser session, or specific engine: google/baidu/sogou)
omb_read	Read a webpage and extract content (supports authenticated pages)
omb_action	Browser interaction: click, fill, scroll, screenshot, evaluate JS, and more
omb_map	Discover interactive elements on a page (filter by label, type, region)
omb_snapshot	Accessibility snapshot with stable element refs (@1, @2...)
omb_guide	Load detailed usage guides for each tool
omb_report	Report page extraction quality issues

Usage Examples

Search the web

omb_search({ query: "OpenClaw plugins", max_results: 5 })

Read a login-required page

omb_read({ url: "https://notion.so/my-private-page" })

Interact with a page

// 1. Read the page
omb_read({ url: "https://example.com" })

// 2. Find elements
omb_map({ tabId: 123, label: "Search", types: ["input"] })

// 3. Fill and submit
omb_action({ action: "fill", tabId: 123, selector: "#search-input", value: "hello" })
omb_action({ action: "keypress", tabId: 123, key: "Enter" })

Take a screenshot

omb_action({ action: "screenshot", tabId: 123, annotate: true })

Troubleshooting

Error	Cause	Solution
Could not find omb binary	omb CLI not installed	Run curl -fsSL https://api.omb.org.cn/install | bash
fetch failed	Browser not running or extension disconnected	Open Chrome, check extension shows "Connected"
AUTH_SESSION_MISSING	Not authorized	Run omb login to authorize
omb process exited	omb CLI crashed	Check ~/.omb/omb exists; reinstall if needed

Requirements

• omb CLI -- installed via the install script above
• Chrome with the Oh My Browser extension installed and running
• Node.js 18+ (for the plugin runtime)

Architecture

Agent --> oh-my-browser plugin --> omb --mcp (subprocess) --> API --> Chrome extension

The plugin itself doesn't call APIs directly. It spawns an omb subprocess and forwards all tool calls via the MCP (stdio JSON-RPC) protocol.

License

MIT