The agent records the flow
It opens your product in a browser, on desktop or as a phone, and does the task step by step. Every click is logged with where it landed. Mistakes and loading time are left out of the video.
An MCP server for AI agents · free
Ask your agent for a training guide, a demo or a launch clip. Latch clicks through your product in a browser, edits the footage and hands back the video, usually in a couple of minutes.
Ask for a bug report and you get a plain recording. Ask for training and you get numbered steps and instructions. Ask for a launch clip and you get titles and transitions. The agent picks the style from what you asked for.
It opens your product in a browser, on desktop or as a phone, and does the task step by step. Every click is logged with where it landed. Mistakes and loading time are left out of the video.
It reads Latch's library, where each style, annotation and transition says when to use it. It writes a timeline, renders a quick preview, checks it and fixes problems before the final render.
The camera zooms toward each click before it happens. Steps get rings, numbers and instructions. Output is 60 fps with motion blur and click sounds, rendered on your machine.
Works with Claude Code, Claude Desktop, Cursor, Codex and any other MCP client. There's no account and no API key.
Each card is a real render. Hover to play it. Click to see the description the agent reads and the line it writes to use it.
Nothing matches that. Try "zoom", "tiktok" or "documentary".