
Point to browser DOM elements for agentic coding tools via MCP!
MCP Pointer is a local tool combining an MCP Server with a Chrome Extension:
- ๐ฅ๏ธ MCP Server (Node.js package) - Bridges between the browser and AI tools via the Model Context Protocol
- ๐ Chrome Extension - Captures DOM element selections in the browser using
Option+Click
The extension lets you visually select DOM elements in the browser, and the MCP server makes this textual context available to agentic coding tools like Claude Code, Cursor, and Windsurf through standardized MCP tools.
- ๐ฏ
Option+Click
Selection - Simply holdOption
(Alt on Windows) and click any element - ๐ Complete Element Data - Text content, CSS classes, positioning, and aggregated CSS rules for the selected subtree
- โ๏ธ React Component Detection - Component names and source files via Fiber (experimental)
- ๐ WebSocket Connection - Real-time communication between browser and AI tools
- ๐ค MCP Compatible - Works with Claude Code and other MCP-enabled AI tools
- ๐ Multiple Element Selection - Select multiple elements in sequence with visual feedback
- ๐ฑ๏ธ Middle Click Selection - Use
Option+Middle Click
to avoid triggering element actions
mcp-pointer-demo.mp4
See MCP Pointer in action: Option+Click
any element in your browser, then ask your agentic coding tool about it (in this example, Claude Code). The AI gets complete textual context about the selected DOM element including CSS properties, url, selector, and more.
MCP Pointer now supports selecting multiple DOM elements in sequence! This feature allows you to build a collection of related elements for comprehensive analysis.
- Enable targeting mode by pressing your trigger key (default varies by system)
- Select first element:
Option+Middle Click
on any element - Select additional elements: Continue
Option+Middle Click
on other elements - Deselect elements:
Option+Middle Click
the same element again to remove it from selection - View selection: All selected elements are highlighted with visual overlays
- Sequential Selection: Elements are numbered in selection order (1, 2, 3...)
- Visual Feedback: Selected elements show distinctive highlighting with position indicators
- Clean Interaction: Uses middle mouse button to avoid triggering element click events
- Dynamic Updates: Selection indexes automatically update when elements are added/removed
- Complete Data: All selected elements are sent together to your AI tool for analysis
- Selection Method:
Option+Middle Click
(prevents accidental element triggering) - Index Display: Sequential numbering with high contrast visibility
- Event Handling: Prevents default click behavior to avoid unwanted page interactions
- Data Structure: Maintains array of selected elements with proper ordering
๐ Now available on Chrome Web Store!
Simply click the link above to install from the Chrome Web Store.
Alternative: Manual Installation
Option A: Download from Releases
- Go to GitHub Releases
- Download
mcp-pointer-chrome-extension.zip
from the latest release - Extract the zip file to a folder on your computer
- Open Chrome โ Settings โ Extensions โ Developer mode (toggle ON)
- Click "Load unpacked" and select the extracted folder
- The MCP Pointer extension should appear in your extensions list
- Reload web pages to activate the extension
Option B: Build from Source
- Clone this repository
- Follow the build instructions in CONTRIBUTING.md
- Open Chrome โ Settings โ Extensions โ Developer mode (toggle ON)
- Click "Load unpacked" and select the
packages/chrome-extension/dist/
folder - Reload web pages to activate the extension
One command setup for your AI tool:
npx -y @mcp-pointer/server config claude # or cursor, windsurf, and others - see below
Other AI Tools & Options
# For other AI tools
npx -y @mcp-pointer/server config cursor # Opens Cursor deeplink for automatic installation
npx -y @mcp-pointer/server config windsurf # Automatically updates Windsurf config file
npx -y @mcp-pointer/server config manual # Shows manual configuration for other tools
Optional: You can install globally with
npm install -g @mcp-pointer/server
to usemcp-pointer
instead ofnpx -y @mcp-pointer/server
After configuration, restart your coding tool to load the MCP connection.
- Navigate to any webpage
Option+Middle Click
any element to select it (middle click prevents triggering element actions)- Ask your AI to analyze the targeted element!
Your AI tool will automatically start the MCP server when needed using the npx -y @mcp-pointer/server start
command.
Available MCP Tool:
get-pointed-element
- Get textual information about the currently pointed DOM element from the browser extension
- Element Selection: Content script captures
Option+Middle Click
events - Data Extraction: Analyzes element structure, CSS, and framework info
- WebSocket Transport: Sends data to MCP server on port 7007
- MCP Protocol: Makes data available to AI tools via MCP tools
- AI Analysis: Your assistant can now see and analyze the element!
- Basic Info: Tag name, ID, classes, text content
- CSS ่งๅ: ๅๅง CSS ๆๆฌ๏ผ่ฆ็้ไธญๅ ็ด ๅๅ ถๅญๅ ็ด ็ๆๆ็ฑป้ๆฉๅจ
- Component Info: React component names and source files (experimental)
- Position: Exact coordinates and dimensions
- Source Hints: File paths and component origins
- โ๏ธ React - Component names and source files via Fiber (experimental)
- ๐ฆ Generic HTML/CSS/JS - Full support for any web content
- ๐ฎ Planned - Vue component detection (PRs appreciated)
- โ Chrome - Full support (tested)
- ๐ก Chromium-based browsers - Should work (Edge, Brave, Arc - load built extension manually)
- Make sure MCP server is running:
npx -y @mcp-pointer/server start
- Check browser console for WebSocket errors
- Verify port 7007 is not blocked by firewall
- Restart your AI assistant after installing
- Check MCP configuration:
mcp-pointer config <your-tool>
- Verify server is running:
npx -y @mcp-pointer/server start
- Some pages block content scripts (chrome://, etc.)
- Try refreshing the page
- Check if targeting is enabled (click extension icon)
- LLM-configurable detail levels (visible text only, all text, CSS levels)
- Progressive refinement options
- Token-conscious data fetching
- Vue.js component detection
- Better React support (React 19 removed
_debugSource
, affecting source mapping in dev builds)
- Base64 encoding for images (img tags)
- Screenshot capture of selected elements
- Separate MCP tool for direct visual content retrieval
MIT License - see LICENSE file for details
We welcome contributions! Please see our CONTRIBUTING.md guide for development setup and guidelines.
Inspired by tools like Click-to-Component for component development workflows.
Made with โค๏ธ for AI-powered web development
Now your AI can analyze any element you point at with Option+Middle Click
! ๐