Comment on a video like a doc. Your agent makes the edits.

Pause the video, drag a box or click a spot, and type what should change. CloudClip saves each comment with its timestamp and pixel coordinates, and your AI agent applies them with ffmpeg.

Try the video demo Try it on an image Install

1Ask your agent

Say "Review ~/Videos/launch.mp4 with CloudClip". The agent starts a local server and opens the review page.

2Pause, mark, comment

Press C, drag a box or click a spot on the frame, and type. Images work the same way, without timestamps.

3Get a new file

The agent grabs each marked frame to see what you pointed at, asks about anything unclear, and writes launch.cloudclip.mp4 next to the original.

What your agent receives

[
  {"time": 72.4, "timestamp": "1:12.4",
   "text": "blur this logo",
   "box": {"x": 1190, "y": 86, "w": 384, "h": 108}},
  {"time": 95.2, "timestamp": "1:35.2",
   "text": "cut from here to 1:50"}
]

Times are in seconds and coordinates are in source pixels, the numbers ffmpeg's crop, boxblur, drawbox and trim filters take.

Install

In Claude Code, desktop app or terminal:

/plugin marketplace add davidpn11/cloudclip
/plugin install cloudclip@cloudclip

It needs python3 and ffmpeg. For Codex, Gemini CLI, Cursor or Grok, clone the repo so the agent reads AGENTS.md, or copy skills/review into the agent's skills folder.