
claude-real-video MCP Server: Let Claude and Cursor Watch Videos
Lets Claude or Cursor watch a video link or file, then search everything it has watched.
This is a community project by HUANGCHIHHUNGLeo. Everything runs on your computer: it downloads the video, picks key frames and transcribes the audio with Whisper. You need ffmpeg installed.
It keeps a cache and a searchable memory of what it watched on your disk.
Good for
- People who want a video summarised or checked frame by frame
- Researchers who keep a searchable record of talks and meetings
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-real-video MCP
Claude Code
- 1
Copy this prompt.
Add the claude-real-video MCP server to Claude Code. Run this command: claude mcp add claude-real-video -- uvx --from "claude-real-video[mcp,whisper]" crv-mcp Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires uv (Python), ffmpeg.
Install ffmpeg first (brew install ffmpeg on macOS, sudo apt install ffmpeg on Linux, winget install Gyan.FFmpeg on Windows). The server is the crv-mcp command from the claude-real-video package, and it needs the mcp extra. The whisper extra adds transcription. The README shows pip install and a bare crv-mcp command; the uvx form here is adapted by us.
Built from the package data we hold. Not yet tested by us.
Tools
Watches a video for you
Downloads a video link or reads a local file, picks the key frames, and transcribes the audio with Whisper on your machine. It returns the transcript plus the first batch of frames as images.
Try: “Watch https://youtu.be/example and summarise what is shown on screen.”
Write ·
watch_videoGets more frames from a watched video
Pages through the remaining key frames of a video that was already analysed.
Try: “Show me the next 12 frames of that video.”
Files ·
get_framesGets just the words of a video
Returns the transcript only, without frames, so a long talk does not use image tokens. Output is capped at 20,000 characters by default.
Try: “Give me the transcript of that talk.”
Files ·
get_transcriptSearches every video you have watched
Searches the transcripts and notes of all videos watched so far to find the one that covered a topic.
Try: “Which video did I watch about database indexing?”
Files ·
search_memoryLists videos already watched
Shows the videos in the local cache so the assistant does not watch the same one twice.
Try: “What videos have you watched so far?”
Files ·
list_watched
Read from the source code
Permissions
medium risk · 2 things to watch
Medium risk
It downloads videos and writes frames, transcripts and a search memory to your disk, and sends frames to your AI app. Avoid videos with private content unless you are fine with that.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It downloads videos and stores them on your disk
watch_video downloads the source and caches frames and transcripts under ~/.cache/crv-mcp. Clear that folder when you are done with sensitive videos. github.com
It reads any local video path you give it
Pass only files you want analysed. The frames are sent to your assistant, so they leave your machine if the assistant is cloud-hosted. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Turning on read-only | No read-only mode. The README says memory can be turned off with CRV_NO_MEMORY=1. |
|---|---|
| Talks to | The video site you give it (downloads through yt-dlp) |
Questions
Is claude-real-video official?
No. It is a community project by HUANGCHIHHUNGLeo and is not made by YouTube or any video site.
Does it need an API key?
No. Transcription runs locally with Whisper and no account is needed.
What do I need to install first?
You need ffmpeg on your machine. The install box has the command, which also adds the transcription (Whisper) extra.
Is my video sent to the cloud?
Analysis runs locally. The frames and transcript do go to your AI app, so they leave your machine if that app is cloud-hosted.
Can I stop it from remembering what it watched?
Yes. The README says the memory can be disabled with the CRV_NO_MEMORY=1 environment variable.
Which videos work?
Links from sites that yt-dlp supports, such as YouTube and Instagram, and local video files.
Information
updated Nov 2, 2026
- Publisher
- Community
- Package
claude-real-video- Registry
- PyPI
- Transport
- stdio
- Auth
- None
- Version
- 0.10.7
- License
- MIT
- Last commit
- Oct 23, 2026
- Open issues
- 1
- Listed in
- Official MCP registry
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 2Hide
- claude-real-video README · github.com
- claude-real-video MCP server source · github.com