Can Claude watch my screen while I show it something?
Not on its own. Claude Code and Claude Desktop can take a single screenshot (computer use), but they can’t watch you do something over time or hear you explain it — users ask for screen sharing in Claude Code’s issue tracker. This records the screen and your voice for the time you choose and hands the AI only the moments the screen changed, with a timestamped transcript. With the GUI Hub Essentials MCP server installed, Claude (Desktop or Code), Cursor, VS Code and other MCP clients can watch your screen live while you show and explain something (screenshots of every change + what you said) locally: the assistant calls the watch_screen tool, which runs FFmpeg + Whisper on your machine. Files are not uploaded.
Set it up (2 minutes)
- Claude Code / Cursor / VS Code / Windsurf: add the MCP server
npx -y https://guihub.cc/mcp.tgz --product=essentials
Claude Code:claude mcp add gui-hub-essentials -- npx -y https://guihub.cc/mcp.tgz --product=essentials - Claude Desktop: download the one-file extension and double-click it.
- No AI? GUI Hub Essentials is also a point-and-click desktop app.
Agent with its own computer?
Grok Bot, ChatGPT dots, Claude Duet, OpenClaw, Manus or Codex can run it directly — no MCP needed:
npx -y https://guihub.cc/cli.tgz run watch_screen --outDir /path/to/folder
Options: npx -y https://guihub.cc/cli.tgz info watch_screen · skill file: SKILL.md · more for agents
Then ask
“Watch my screen for a minute while I reproduce the checkout bug, then find the cause in the code.”
What the AI gets
watch_screen — Watch the user's screen live: records the screen (and microphone, when available) for `seconds`, while the user shows you something and explains it out loud — reproducing a bug, a task to automate, how an app is used. Saves into `outDir`: frame_<N>.jpg each time the screen changes (N = tenths of a second from the start, so frame_123.jpg = 12.3 s; only changed screens, at most 300), transcript.txt with [hh:mm:ss] lines of what the user said, and recording.mp4. BEFORE calling, tell the user: 'Recording your screen for N seconds starts now — show me and talk me through it.' Afterwards open transcript.txt and the frames to understand what happened. macOS needs Screen Recording (and Microphone) permission for the app that runs the MCP server; Linux needs an X11 session (not Wayland).
| Input | Type |
|---|---|
| seconds | string (20, 60, 120, 300) |
| mic | string (yes, ) |
| language | string (, en, es, fr, de, it) |
| outDir* | string |
Cost
Free to install; each tool has 3 free uses. Then $9 once for all of GUI Hub Essentials (no subscription). The tool tells the assistant when free uses run out and how to unlock.
Powered by FFmpeg + Whisper (GPL (FFmpeg static build), Apache-2.0/MIT (Whisper via sherpa-onnx)), the official release, verified by SHA-256. GUI Hub is independent and not affiliated with FFmpeg + Whisper, Anthropic, Cursor or Microsoft.
Other things your AI can do with GUI Hub
- Can Claude make a video with a voiceover from my photos?
- Can Claude turn a long video into Shorts, Reels or TikToks?
- Can Claude find duplicate files on my computer?
- Can Claude search my photos for something?
- Can Claude find a specific photo on my computer?
- Can Claude search inside my PDFs and Word documents?
- Can Claude search my voice memos?
- Can Claude search my old emails from a Gmail export?
- Can Claude find a document on my computer by what it says?
- Can Claude read a PDF that is on my computer?
- Can Claude read and search my scanned PDFs?
- Can Claude turn photos of a document into a searchable PDF?