mcp-pointer

Why we included this project

Agentic coding tools are good at reading code but blind to the page you're looking at. MCP Pointer closes that gap with a small MCP server and a Chrome extension: hold Option and click any element, and the extension captures its text, classes, attributes, position, and styling, then sends that structured context to your assistant over a WebSocket. The AI ends up with the same DOM you see, so it can target selectors, debug layout, or edit markup without you pasting HTML. Per-call controls let you request visible-only text, suppress text entirely, or dial CSS detail from none to full computed styles, which keeps prompts lean. There's also experimental React component detection that surfaces component names and source files, and for frontend work with Claude Code, Cursor, or Windsurf, this is a practical bridge between what you see and what the agent understands.