webagent-mcp
v1.1.0
Published
MCP server for browser automation - give AI agents the power to browse the web
Maintainers
Readme
WebMCP
MCP Server for browser automation via Chrome extension
WebMCP enables AI agents (Claude, Cursor, etc.) to control your browser through the Model Context Protocol.
Quick Start
1. Install the MCP Server
npm install -g webagent-mcpOr run directly:
npx webagent-mcp2. Install Chrome Extension
Load the extension in Chrome:
- Go to
chrome://extensions - Enable "Developer mode"
- Click "Load unpacked"
- Select the
webagent-extensionfolder
3. Configure Your AI Client
Claude Desktop
Add to ~/.config/claude/claude_desktop_config.json:
{
"mcpServers": {
"webagent": {
"command": "npx",
"args": ["webagent-mcp"]
}
}
}Cursor
Add to MCP settings in Cursor preferences.
Available Tools
| Tool | Description |
|------|-------------|
| web_navigate | Navigate to a URL |
| web_read_page | Read page content, forms, buttons |
| web_click | Click elements |
| web_type | Type text into inputs |
| web_select | Select dropdown options |
| web_scroll | Scroll page up/down |
| web_press_key | Press keyboard keys |
| web_analyze_form | Analyze form structure |
| web_fill_form | Fill form intelligently |
| web_list_tabs | List open browser tabs |
| web_create_tab | Open new tab |
| web_close_tab | Close a tab |
| web_screenshot | Capture visible area |
| web_wait | Wait for specified time |
Configuration
Environment variables:
| Variable | Default | Description |
|----------|---------|-------------|
| WEBMCP_WS_PORT | 8080 | WebSocket port for extension |
| WEBMCP_SSE_PORT | 3000 | SSE port for HTTP mode |
| WEBMCP_TOOL_TIMEOUT | 60000 | Tool execution timeout (ms) |
| WEBMCP_RETRY_ATTEMPTS | 3 | Retry attempts on failure |
| WEBMCP_RATE_LIMIT | 60 | Max actions per minute |
| WEBAGENT_MOCK | false | Enable mock mode |
Usage Examples
Navigate and Read
User: Go to example.com and tell me what you see
AI uses:
1. web_navigate { url: "https://example.com" }
2. web_read_page {}Fill a Form
User: Fill the contact form with my info
AI uses:
1. web_read_page {} → sees form structure
2. web_click { selector: "input[name='email']" }
3. web_type { selector: "input[name='email']", text: "[email protected]" }
4. web_click { selector: "button[type='submit']" }Development
# Clone repo
git clone https://github.com/user/webmcp.git
# Install dependencies
cd webagent-mcp
npm install
# Run in development
npm run dev
# Run tests
npm testLicense
MIT
