Browser-Use MCP Console

🤖 智能浏览器自动化 + 控制台调试工具

A powerful MCP server that enables AI assistants to control browsers and debug web applications by capturing console outputs in real-time.

✨ Key Features

• 🌐 Browser Automation - AI-powered web automation using LLMs
• 🐛 Console Monitoring - Automatic capture of JavaScript errors, warnings, and logs
• ⚡ Parallel Execution - Run multiple browser tasks simultaneously
• 📝 Error Reporting - Detailed console error messages in task results
• 🔧 Debug Support - Perfect for testing and debugging web applications

🚀 Quick Start

1. Install the Package

pip install browser-use-mcp-console

Or use with Chinese mirror:

pip install browser-use-mcp-console -i https://mirrors.aliyun.com/pypi/simple/

2. Get Your API Key

Choose one provider:

• OpenRouter - Recommended, supports multiple models
• OpenAI - GPT models
• Anthropic - Claude models

3. Configure with Claude Desktop

# Using OpenRouter (Recommended)
claude mcp add browser-use \
  --scope user \
  --command "uvx" \
  --args "browser-use-mcp-console" \
  --env OPENROUTER_API_KEY=your-api-key-here

# Or using OpenAI
claude mcp add browser-use \
  --scope user \
  --command "uvx" \
  --args "browser-use-mcp-console" \
  --env OPENAI_API_KEY=your-api-key-here

4. Configure with Cursor (Optional)

Add to Cursor's MCP settings:

{
  "mcpServers": {
    "browser-use": {
      "command": "uvx",
      "args": ["browser-use-mcp-console"],
      "env": {
        "OPENROUTER_API_KEY": "your-api-key-here"
      }
    }
  }
}

📖 Usage Examples

After configuration, you can use natural language commands in Claude:

Basic Web Automation

"Open Baidu Translate and translate 'Hello' to Chinese"
"Login to website with username admin and password 123456"
"Fill out the contact form on example.com"

Debug & Testing

"Visit localhost:3000 and check for console errors"
"Test the login functionality and report any JavaScript errors"
"Click the submit button and monitor console output"

Parallel Tasks

"Open three websites simultaneously: google.com, baidu.com, bing.com"
"Run multiple tests: login test, form validation test, API test"

⚙️ Configuration Parameters

Parameter	Type	Default	Description
tasks	list[str]	Required	Task descriptions (single or multiple)
model	str	Varies*	LLM model to use
headless	bool	False	Run browser in background
max_steps	int	100	Maximum actions per task
use_vision	bool	True	Enable visual page understanding

*Default models:

• OpenRouter: google/gemini-2.5-pro
• OpenAI: gpt-4o-mini
• Anthropic: claude-3-5-sonnet-20241022

🔍 Console Monitoring Features

The key advantage of this MCP server is automatic console capture:

• Automatic Error Detection - Captures all JavaScript errors and unhandled Promise rejections
• Detailed Error Reports - Returns specific error messages with stack traces
• Real-time Monitoring - Console logs are captured as pages execute
• Comprehensive Coverage - Monitors console.log, console.error, console.warn, console.info

Example error report:

Task completed. Found console errors:
- TypeError: Cannot read properties of undefined (reading 'data') at app.js:45
- Unhandled Promise rejection: API timeout at fetch.js:89

🛠️ Troubleshooting

Browser not opening?

• The server auto-detects Playwright browsers
• First run may download Chromium automatically
• Check PLAYWRIGHT_BROWSERS_PATH environment variable

API key issues?

• Ensure your API key is valid and has credits
• OpenRouter users: Check your balance at openrouter.ai
• Try testing with a simple task first

Console logs not showing?

• Console capture is automatic but viewing requires explicit checking
• The agent will check console when task mentions "debug", "test", or "errors"

📚 Development

# Clone repository
git clone https://github.com/yourusername/browser-use-mcp-console
cd browser-use-mcp-console

# Install in development mode
pip install -e .

# Run tests
python test_mcp_client.py

📄 License

MIT License