
Claude Video Vision: Let Claude Watch Videos
Lets Claude Code watch your videos and YouTube links by looking at frames and reading a transcript.
This is a community plugin for Claude Code. Its server extracts still frames with ffmpeg and gets a transcript of the audio. It does the looking and listening, and Claude does the interpreting.
You pick how audio is handled. The local Whisper option runs on your computer. The Gemini and OpenAI options send the audio to those companies and need your own API key. You also need ffmpeg installed and Node.js 24 or later.
Good for
- Developers who want Claude to look at a screen recording of a bug
- People who need a summary of a long talk or tutorial
- Anyone who wants Claude to read the text on screen at a given moment
Works in
12 should work
Tap an app to see how to set it up and any known problems.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Client does not support this transport.
Client does not support this transport.
Install claude-video-vision MCP
Claude Code
- 1
In Claude Code, run /plugin marketplace add https://github.com/jordanrendric/claude-video-vision
- 2
Run /plugin install claude-video-vision (as a separate command).
- 3
Run /claude-video-vision:setup-video-vision to choose a backend.
Requires Node.js 24+, ffmpeg.
Read-only mode: There is no read-only mode.
Install ffmpeg first, and Node.js 24 or later. The plugin starts the server for you with npx the first time, then the setup command picks your audio backend. The terminal command and project config here are our adaptation for people who skip the plugin.
From the project README. Source.
Tools
Watches a video and returns frames and a transcript
Takes a video file or YouTube link, pulls still frames with ffmpeg, and gets a timed transcript of the audio. Claude sees the frames as images and reads the words with timestamps. YouTube links are downloaded first.
Try: “Watch ~/Downloads/demo.mp4 and tell me what the bug is”
Runs code ·
video_watchMaps a long video before you look closer
Checks the video's structure with ffmpeg: scene changes, silent stretches and motion. It can also transcribe the audio. It extracts no frames, so Claude can plan which parts to look at.
Try: “Analyze this one-hour lecture and find where the topic changes”
Runs code ·
video_analyzeLooks closely at chosen parts of a video
Pulls frames from the time ranges you pick, at the frame rate and size you choose for each. With the optional index on, it keeps frames cached so repeat looks are quick.
Try: “Look closely at 1:25 to 1:35 and read the text on screen”
Runs code ·
video_detailShows a video's length, size and format
Returns metadata such as duration, resolution and codec for a file or YouTube link, without processing it.
Try: “How long is this video and what resolution is it?”
Runs code ·
video_infoChecks that the needed programs are installed
Looks for ffmpeg, the Whisper engine, the Gemini key and the optional yt-dlp, and tells you what is missing and how to install it.
Try: “Check whether video watching is set up correctly”
Runs code ·
video_setupChanges video settings, and can clear saved sessions
Sets the audio backend, Whisper model, frame size, frame rate and similar options in the config file. One option also deletes the saved session cache folder.
Try: “Switch the audio backend to local Whisper and use 720-pixel frames”
Delete ·
video_configure
Read from the source code
Permissions
medium risk · 5 things to watch
Medium risk
The server runs ffmpeg and other programs on your computer and can upload audio to Google or OpenAI if you choose those backends. Use the local Whisper backend for private videos.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
Gemini and OpenAI backends upload your audio
With either cloud backend, the audio from your video goes to that company under your API key. Pick the local backend for private videos. github.com
It runs programs on your computer
The server starts ffmpeg, and yt-dlp and Whisper when you use them, and downloads Whisper models with curl. Install those tools from sources you trust. github.com
One setting deletes the saved session cache
The clear-sessions option in video_configure removes the sessions folder under ~/.claude-video-vision. It does not touch your videos. github.com
YouTube downloads and models take disk space
Downloaded videos and Whisper models are stored under ~/.claude-video-vision. Old downloads and sessions are cleaned up after 7 days by default. github.com
The tool descriptions tell Claude how to order calls
The watch tool says to call the analyze tool first for videos longer than 30 seconds. This changes how many calls Claude makes. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Talks to | Google Gemini API or OpenAI API, only if you pick that audio backend. huggingface.co (one-time Whisper model download for the local backend). YouTube, through yt-dlp, for links |
|---|---|
| Sign-in and keys | No sign-in. API keys, if used, come from the GEMINI_API_KEY or OPENAI_API_KEY environment variables. |
Questions
Is Claude Video Vision official?
No. It is a community project by Jordan Vasconcelos, not made by Anthropic. It is MIT licensed.
What do I need installed?
Node.js 24 or later and ffmpeg. YouTube links also need yt-dlp. The local audio backend needs whisper.cpp or Python Whisper, and the model downloads itself on first use.
Does it work outside Claude Code?
It is built as a Claude Code plugin with a slash command and a skill. The server is a standard stdio server on npm, so other apps can run it, but the docs only cover Claude Code.
Is my video uploaded anywhere?
Frames go to Claude as images in your conversation. Audio stays on your computer with the local backend. With the Gemini or OpenAI backend, the audio is sent to that company.
Which audio backend should I pick?
Local Whisper is free and offline. Gemini has a free tier of 1,500 requests a day and picks up non-speech sounds. OpenAI Whisper is paid per use.
Can it watch YouTube videos?
Yes, if yt-dlp is installed. It uses English subtitles when they exist, then automatic captions, and falls back to your audio backend when captions are missing.
Why does it fail on Windows?
An old npm can mishandle the package name and print an unknown command error. Update npm or reinstall Node.js 24 or later. The README also gives a global install workaround.
Can I make it read-only?
No. It has no read-only mode. It reads your videos, but it also writes temporary files and its config, and one option clears its session cache.
Information
updated Nov 2, 2026
- Publisher
- Community
- Package
claude-video-vision- Registry
- npm
- Transport
- stdio
- Auth
- None
- Version
- 1.4.0
- License
- MIT
- Last commit
- Oct 14, 2026
- Open issues
- 3
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 3Hide
- Claude Video Vision README · github.com
- Claude Video Vision privacy notes · github.com
- Tool registrations at v1.4.0 · github.com