English · 简体中文 · 日本語 · Español · Português

browserwing.com

Turn any website into structured data with one command. Browser automation platform with 78 built-in scripts, CLI for AI agents, MCP & Skills protocol support, and a visual recorder.

# Install and try it in 30 seconds
npm install -g browserwing && browserwing --port 8080

# Get data from any supported site
browserwing run github-trending
browserwing run bilibili-hot
browserwing run hackernews-top | jq '.[0:5]'

Native Browser Automation Platform with AI Integration

• Complete Browser Control: 26+ HTTP API endpoints for full-featured browser automation
• Built-in AI Agent: Direct conversational interface for browser automation tasks
• Universal AI Tool Integration: Native MCP & Skills protocol support - compatible with any AI tool that supports these standards
• Visual Script Recording: Record browser actions, edit visually, and replay with precision
• Flexible Export Options: Convert recorded scripts to MCP commands or Skills files for AI tool integration
• Intelligent Data Extraction: LLM-powered semantic extraction supporting OpenAI, Claude, DeepSeek, and more
• Session Management: Robust cookie and storage handling for stable, authenticated browsing sessions

• Google Chrome or Chromium installed and accessible in your environment.

You can now combine any scripts into a SKILL.md.

Simply send the following message to your AI agent (OpenClaw, Cursor, Claude Code, etc.):

"Help me install BrowserWing following https://raw.githubusercontent.com/browserwing/browserwing/main/INSTALL.md"

The agent will read the guide and handle the installation, configuration, Chrome setup, and Skill integration automatically.

Using npm:

npm install -g browserwing
browserwing --port 8080

Using pnpm:

pnpm add -g browserwing
browserwing --port 8080

The npm package automatically tests GitHub and Gitee mirrors during installation and selects the fastest one.

⚠️ macOS Users:
If you encounter a "killed" error when running, fix it with:

xattr -d com.apple.quarantine $(which browserwing)

See the macOS Installation Fix Guide for details.

Using Homebrew (macOS/Linux):

# Coming soon
brew install browserwing

Linux / macOS:

curl -fsSL https://raw.githubusercontent.com/browserwing/browserwing/main/install.sh | bash

Windows (PowerShell):

iwr -useb https://raw.githubusercontent.com/browserwing/browserwing/main/install.ps1 | iex

The script automatically:

• Detects your OS/architecture
• Tests GitHub and Gitee mirrors, selects the fastest one
• Downloads and extracts the binary
• Adds to PATH

Then start BrowserWing:

browserwing --port 8080
# Open http://localhost:8080 in your browser

Note for users in China: The installation script automatically uses Gitee mirror if GitHub is slow.

Download the prebuilt binary for your OS from Releases:

# Linux/macOS
chmod +x ./browserwing
./browserwing --port 8080

# Windows (PowerShell)
./browserwing.exe --port 8080

# Install deps (Go + pnpm required)
make install

# Build integrated binary (frontend embedded)
make build-embedded
./build/browserwing --port 8080

# Or build all targets and packages
make build-all
make package

Three Ways to Use BrowserWing:

Configure BrowserWing as an MCP server in any MCP-compatible AI tool:

{
  "mcpServers": {
    "browserwing": {
      "type": "http",
      "url": "http://localhost:8080/api/v1/mcp/message"
    }
  }
}

Paste this configuration into your AI tool's MCP settings to enable browser automation capabilities.

Download and import the Skills file into any AI tool that supports the Skills protocol:

1. Start BrowserWing
2. Download SKILL.md from the repository
3. Import into your AI tool's Skills settings
4. Start automating with natural language commands

Example:

"Navigate to example.com, search for 'AI tools', and extract the top 5 results"

BrowserWing includes a CLI designed for AI agents and shell pipelines. Run scripts and get structured data directly from the terminal — no browser window needed (headless by default).

# List all available scripts
browserwing ls --format=json

# Run a script and get JSON output
browserwing run bilibili-hot
browserwing run github-trending
browserwing run hackernews-top

# Pass parameters to scripts
browserwing run jd-search --keyword="机械键盘"

# Pipe to other tools
browserwing run sinafinance-rank --format=csv > stocks.csv
browserwing run hackernews-top | jq '.[0:5]'

# Show browser window for debugging
browserwing run zhihu-hot --no-headless

78 built-in scripts across 10 categories (Bilibili, GitHub, Reddit, Hacker News, YouTube, Steam, BBC, Bloomberg, Reuters, Google Scholar, Binance, Amazon, Weibo, Zhihu, CNKI, Yahoo Finance, etc.) — ready to use out of the box.

AI Agent Workflow:

1. browserwing ls --format=json — discover available scripts and their parameters
2. browserwing run <name> --key=value — execute and get structured JSON data
3. Parse the output for further processing

Use BrowserWing's built-in AI Agent for immediate browser automation:

1. Open BrowserWing web interface at http://localhost:8080
2. Navigate to "AI Agent" section
3. Configure your LLM (OpenAI, Claude, DeepSeek, etc.)
4. Start conversational browser automation

Export Custom Scripts:

# Export your recorded scripts as Skills or MCP commands
curl -X POST 'http://localhost:8080/api/v1/scripts/export/skill' \
  -H 'Content-Type: application/json' \
  -d '{"script_ids": []}' \
  -o MY_CUSTOM_SCRIPTS.md

Professional Browser Automation with AI Integration

• Universal Protocol Support: Native MCP & Skills implementation works with any compatible AI tool
• Complete Automation API: 26+ HTTP endpoints providing comprehensive browser control capabilities
• Flexible Integration Options: Use as MCP server, Skills file, or standalone AI Agent
• Visual Workflow Builder: Record, edit, and replay browser actions without writing code
• Token-Efficient Design: Optimized for LLM usage with fast performance and minimal token consumption
• Production-Ready: Stable session management, cookie handling, and error recovery
• Extensible Architecture: Convert recorded scripts to reusable MCP commands or Skills files
• Multi-LLM Support: Works with OpenAI, Anthropic, DeepSeek, and other providers
• Enterprise Use Cases: Data extraction, RPA, testing, monitoring, and agent-driven automation

1. Choose Integration Method

   • Copy MCP server configuration for AI tool integration
   • Download Skills file for Skills-compatible AI tools
   • Or use built-in AI Agent for immediate access

2. Configure Your AI Tool

   • Import MCP configuration or Skills file into your preferred AI tool
   • Configure LLM settings (API keys, model selection)
   • Verify connection to BrowserWing

3. Start Automating

   • Control browser through natural language commands
   • Record custom scripts for repeated tasks
   • Export scripts as MCP commands or Skills for reuse

For Browser Automation:

• Launch and manage multiple browser instances
• Configure profiles, proxies, and browser settings
• Handle cookies and authentication sessions
• Execute complex interaction sequences

For Script Recording:

• Capture clicks, inputs, navigation, and waits
• Edit actions visually in the script editor
• Test and debug with step-by-step replay
• Add variables and conditional logic

For AI Integration:

• Convert scripts to MCP commands or Skills files
• Integrate with multiple LLM providers
• Use semantic extraction for data parsing
• Build agent-driven automation workflows

BrowserWing supports CloakBrowser — a stealth Chromium binary with source-level fingerprint patches that passes all major bot detection tests.

Why CloakBrowser?

• 49-57 C++ source-level patches (canvas, WebGL, audio, fonts, GPU, WebRTC, etc.)
• Passes Cloudflare Turnstile, FingerprintJS, BrowserScan, reCAPTCHA v3
• Works as a drop-in replacement for standard Chromium

Setup:

1. Install CloakBrowser:

pip install cloakbrowser
python -c "from cloakbrowser import ensure_binary; ensure_binary()"

2. Launch CloakBrowser's cloakserve CDP server:

python /path/to/cloakbrowser/bin/cloakserve --port=9222 --headless=false

3. Configure BrowserWing to use CloakBrowser's CDP endpoint. Add to your config.toml:

[browser]
ControlURL = "http://localhost:9222"

Or set the BROWSER_CONTROL_URL environment variable:

export BROWSER_CONTROL_URL="http://localhost:9222"
browserwing --port 8080

4. Important: CloakBrowser handles fingerprint spoofing at the binary level, so BrowserWing's JS-level stealth mode will conflict with CloakBrowser's C++ patches. When using CloakBrowser, set use_stealth = false in your browser config or the USE_STEALTH environment variable to disable BrowserWing's JS stealth injection.

How it works:

• cloakserve is a CDP multiplexer that spawns one Chrome process per fingerprint seed
• Each Chrome process has compiled-in fingerprint patches (no JS injection needed)
• BrowserWing connects to cloakserve via HTTP, resolves the WebSocket URL automatically, and controls the browser via standard CDP
• Cookie persistence and session management work the same as with regular Chrome

Key benefits:

• Stronger bot detection evasion than JS-level stealth injection
• Consistent fingerprints across sessions (seed-based)
• Native SOCKS5 proxy and geoip-based timezone/locale detection built-in
• Works with Playwright, Puppeteer, LangChain, browser-use, and BrowserWing

BrowserWing exposes 26+ RESTful endpoints for programmatic browser control:

Navigation & Control

• Navigate to URLs, go back/forward, refresh pages
• Manage browser windows and tabs
• Handle page loading and timeouts

Element Interaction

• Click, type, select, and hover actions
• File uploads and form submissions
• Keyboard shortcuts and key presses

Data Extraction

• Extract text, HTML, and attributes
• Semantic content analysis with LLM
• Screenshot capture (full page or element)

Advanced Operations

• Execute custom JavaScript
• Manage cookies and local storage
• Batch operations for efficiency
• Wait conditions and element visibility

Complete Documentation: See docs/EXECUTOR_HTTP_API.md for detailed endpoint specifications

For AI agent integration, the CLI uses structured exit codes:

Set BROWSERWING_JSON_ERRORS=1 for machine-readable JSON error output on stderr.

• Issues and PRs are welcome. Please include clear steps to reproduce or a concise rationale.
• For feature ideas, open a discussion with use cases and expected outcomes.

Discord: https://discord.gg/BkqcApRj twitter: https://x.com/chg80333

• Inspired by modern browser automation, agentic workflows, and MCP.

• MIT License. See LICENSE.

• Do not use for illegal purposes or to violate site terms.
• Intended for personal learning and legitimate automation only.