BrowserAgent MCP 🌐

Give AI the ability to browse the web. 37 automation tools for Claude Desktop.

What It Does

BrowserAgent MCP is a Model Context Protocol server that gives AI agents real browser control:

• 🌐 Navigate to any website
• 📸 Screenshot pages
• 🖱️ Click buttons and links
• ⌨️ Type into forms
• 📜 Scroll and wait for elements
• 🤖 AI-powered element finding (optional)

Quick Start

1. Install

npm install -g @vivek3/browser-agent-mcp
npx playwright install chromium

2. Configure Claude Desktop

Add to your claude_desktop_config.json:

Windows: %APPDATA%\Claude\claude_desktop_config.json
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "browser-agent": {
      "command": "npx",
      "args": ["@vivek3/browser-agent-mcp"]
    }
  }
}

3. Use It

Restart Claude Desktop and try:

• "Go to news.ycombinator.com and tell me the top 3 stories"
• "Take a screenshot of google.com"
• "Fill out this form..."

All 37 Tools

Core

| Tool | Description | |------|-------------| | browser_navigate | Go to a URL | | browser_screenshot | Capture the page | | browser_click | Click an element | | browser_type | Type text | | browser_get_content | Read page text |

Navigation

| Tool | Description | |------|-------------| | browser_scroll | Scroll up/down | | browser_wait | Wait for element | | browser_back | Go back | | browser_forward | Go forward | | browser_get_url | Get current URL | | browser_get_title | Get page title |

Forms & Interaction

| Tool | Description | |------|-------------| | browser_hover | Hover over element | | browser_select | Choose from dropdown | | browser_check | Toggle checkbox | | browser_upload | Upload file | | browser_eval | Run JavaScript | | browser_get_links | Extract all URLs | | browser_get_html | Get HTML source |

AI-Enhanced (requires Gemini API key)

| Tool | Description | |------|-------------| | browser_smart_click | Click by description | | browser_smart_type | Type by field description | | browser_find_element | Find by natural language | | browser_summarize | AI page summary | | browser_analyze_screenshot | Visual analysis |

Session & Tabs

| Tool | Description | |------|-------------| | browser_get_cookies | Get cookies | | browser_set_cookies | Set cookies | | browser_clear_cookies | Clear cookies | | browser_get_storage | Get localStorage | | browser_set_storage | Set localStorage | | browser_new_tab | Open new tab | | browser_switch_tab | Switch tabs | | browser_close_tab | Close tab | | browser_list_tabs | List all tabs | | browser_save_pdf | Save as PDF |

Monitoring

| Tool | Description | |------|-------------| | browser_capture_requests_start | Monitor network | | browser_get_requests | Get network log | | browser_console_start | Monitor console | | browser_get_console | Get console log | | browser_get_performance | Core Web Vitals | | browser_a11y_check | Accessibility audit |

AI Features (Multiple LLM Options)

Set one of these environment variables:

| Provider | Environment Variable | Free Tier? | |----------|---------------------|------------| | Google Gemini | GEMINI_API_KEY | ✅ Yes | | OpenAI GPT-4 | OPENAI_API_KEY | ❌ Paid | | Anthropic Claude | ANTHROPIC_API_KEY | ❌ Paid | | Ollama (Local) | LLM_PROVIDER=ollama | ✅ Free (local) |

# Example: Use Gemini (free)
set GEMINI_API_KEY=your_key_here

# Example: Use OpenAI
set OPENAI_API_KEY=sk-...

# Example: Use local Ollama
set LLM_PROVIDER=ollama
# Make sure Ollama is running: ollama serve

Auto-detection: If multiple keys are set, priority is: Gemini → OpenAI → Anthropic → Ollama

Security

See SECURITY.md for security policy and considerations.

Key points:

• browser_eval runs in browser sandbox (safe)
• No data is logged or stored
• API keys read from environment only

License

MIT © vv-vivek

⭐ Star this repo if you find it useful!