Skip to content
Hoh JDLatest guide
All guides
AI tools10 min setup

Watch for Claude Code: Beginner Setup Guide

Jump to a section

Teach Claude to watch any video. You paste in a YouTube, Instagram or TikTok link (or a video file), and Claude gets the frames and the transcript, so it can tell you what happens in the video.

What you need

  1. The Claude Desktop app. Download it from https://claude.ai/download and sign in.
  2. A paid Claude plan that includes Claude Code (Pro or Max).
  3. A Mac or a Windows PC. This guide gives the Mac steps first and Windows steps where they differ.

Watch only works in Claude Code, which is the Code tab in the desktop app. It does not work in normal Claude Chat or Cowork.

Step 1: Install the Watch plugin in Claude Desktop

  1. Open Claude Desktop and click Customize in the left sidebar.
  2. Select Plugins, click Add, then choose Add marketplace.
  3. Choose Add from a repository.
  4. In the URL box, paste the address below and click Sync:
    https://github.com/bradautomates/claude-video
  5. Find Watch in the marketplace you just added and click Install.
    Adding the marketplace alone doesn't install Watch, so make sure you click Install.

If you don't see Customize or Plugins, update Claude Desktop to the latest version.

Step 2: Install the tools Watch needs

Watch uses three free programs: Python, FFmpeg (handles video) and yt-dlp (downloads videos).

The easy way: let Claude do it

Open a Code session in Claude Desktop and paste this:

Use the watch skill's bundled setup.py to check dependencies in this session.
Tell me which tools are missing and help me install them here.

On a Mac, Claude can usually install whatever is missing for you.

The manual way

Mac

  1. Open the Terminal app.
  2. If you don't have Homebrew yet, install it from https://brew.sh (copy the one-line command on that page into Terminal).
  3. Then run:
    brew install python ffmpeg yt-dlp

Windows

  1. Install Python 3.10 or newer from https://www.python.org/downloads/windows/ and tick "Add Python to PATH" during setup.
  2. Open PowerShell and run these one at a time:
    winget install --id Gyan.FFmpeg --exact
winget install --id yt-dlp.yt-dlp --exact
winget install --id DenoLand.Deno --exact

When you're done, quit Claude Desktop completely and reopen it so it can find the new programs.

Step 3 (optional but recommended): Add a free Gemini key

This is the fastest and most accurate option. Google's video model watches the whole video, picture and sound together, and sends the answer back to Claude.

  1. Get a free key at https://aistudio.google.com/apikey.
  2. The first time you use Watch, its setup wizard asks for the key. Paste it in when asked.
    You can also have Claude open the file ~/.config/watch/.env, then paste your key after GEMINI_API_KEY= and save it yourself.

No key? That's fine. Choose local when asked. Watch then pulls frames and captions on your own computer instead. Use local for private videos you don't want uploaded to Google.

Keep your key private. Don't share it in screenshots or public posts.

Step 4: Watch your first video

  1. In Claude Desktop, start a new Code session.
  2. Type /watch and choose /watch:watch from the menu, or just ask Claude to "use the watch skill".
  3. Paste a video link and say what you want. For example:
/watch https://www.youtube.com/watch?v=VIDEO_ID
Summarize this video and give me the key timestamps.

The first run walks you through a short setup. If you're unsure, pick balanced detail and none for transcription.

What success looks like: Claude replies with a summary that includes timestamps.

Prompts to copy

Summarize a long YouTube tutorial

/watch <youtube link>
I don't have time to watch this. Summarize everything, break it down
step by step, and list the most important timestamps.

Break down a viral Reel or TikTok

/watch <reel or tiktok link>
Analyze exactly why this video works: the hook, the storytelling and
the editing structure (cuts, text on screen, pacing). Put the breakdown
in an Excel spreadsheet with columns: Time, Beat, What happens, Why it works.

Turn it into your own script

Using that breakdown, write a similar script for my own content about <your topic>.
Keep the same structure: hook, problem, demo, payoff, call to action.

Focus on one part of a long video

/watch <link> --start 12:00 --end 15:30
What exactly do they do in this section? Explain it step by step.

Read text on screen (code, slides, charts)

/watch <link> --resolution 1024
Write down the commands shown on screen.

Tips

  • Shorter videos work best. Under 10 minutes is ideal. For long videos, ask about a section with --start and --end.
  • Instagram and TikTok sometimes need you to be logged in. If a link fails, download the video and give Claude the file instead, for example: /watch my-video.mp4.
  • Local files: put the video in the folder your Code session is working in, then refer to it by name.
  • Keep yt-dlp updated, because sites change often. On a Mac run brew upgrade yt-dlp. On Windows run winget upgrade --id yt-dlp.yt-dlp --exact.

Troubleshooting

Problem Fix
/watch doesn't show up Make sure you clicked Install on Watch, not just added the marketplace. Start a new Code session.
"Command not found", or FFmpeg/yt-dlp is missing Do Step 2 again, then quit and reopen Claude Desktop.
403 error or a login challenge Update yt-dlp (see Tips) and try again. For private videos, download the file and use that.
Gemini error (auth, quota or upload) Check your key in ~/.config/watch/.env, wait a minute if you hit a rate limit, or run it again with --engine local.
Windows opens the Microsoft Store when you type python Reinstall Python from python.org with "Add Python to PATH" ticked.
It works in Code but not in Chat That's expected. Watch only works in the Code tab.

How to update

In Claude Desktop, go to Customize → Plugins → Add → Manage marketplaces, select claude-video, and update it. Then start a new Code session.

Watch for Claude Code is made by bradautomates and released under the MIT license.