Peek

Point at a page element and send it — selector, DOM, and a cropped screenshot — to your coding agent.

Point at an element in your real browser and hand it to your coding agent. Click the Peek icon, click the elements you care about, hit Done. Peek copies a prompt containing a unique CSS selector, the XPath, the element HTML, its box and computed styles, and a cropped screenshot. Paste that into Claude Code, Codex, Cursor, or Grok Build and say what to change. This is not a browser-driving agent. You pick. The model looks. WHY IT HELPS - Stop describing which button is broken. Point at it. - The agent gets a real selector, so it can find the component in your code. - The cropped screenshot shows the visual bug instead of the whole page. - Pin several elements in one pass and send them together. HOW IT WORKS Peek talks to a small helper process on your own machine over 127.0.0.1. The helper registers an MCP server and an agent skill, so your agent can read the pinned elements directly instead of you pasting screenshots around. REQUIRES THE PEEK CLI curl -fsSL https://raw.githubusercontent.com/iammayron/peek/main/install.sh | sh Source and docs: https://github.com/iammayron/peek PRIVACY Everything stays on localhost. Pinned elements are written to ~/.peek on your own machine with owner-only permissions. No accounts, no analytics, no telemetry, and nothing is sent to any server.