#259259th of 324 in Skills, plugins and prompts

claude-video-vision

Claude Code plugin that lets Claude watch videos via ffmpeg frames and transcripts

Stars
1.4k
License
MIT
Last commit
Oct 2026
Last release
Oct 2026
Language
TypeScript
Stars, last 30 weeks+1,444
3 Apr 2026: 2 stars10 Apr 2026: 2 stars17 Apr 2026: 2 stars24 Apr 2026: 200 stars1 May 2026: 420 stars8 May 2026: 534 stars15 May 2026: 600 stars22 May 2026: 661 stars29 May 2026: 693 stars5 Jun 2026: 743 stars12 Jun 2026: 779 stars19 Jun 2026: 803 stars26 Jun 2026: 827 stars3 Jul 2026: 854 stars10 Jul 2026: 960 stars17 Jul 2026: 999 stars24 Jul 2026: 1,034 stars31 Jul 2026: 1,059 stars7 Aug 2026: 1,139 stars14 Aug 2026: 1,216 stars21 Aug 2026: 1,246 stars28 Aug 2026: 1,269 stars4 Sept 2026: 1,283 stars11 Sept 2026: 1,291 stars18 Sept 2026: 1,305 stars25 Sept 2026: 1,325 stars2 Oct 2026: 1,337 stars9 Oct 2026: 1,353 stars16 Oct 2026: 1,377 stars23 Oct 2026: 1,413 stars30 Oct 2026: 1,446 stars

2 on 3 Apr 20261,446 on 30 Oct 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

An MCP server and Claude Code plugin that extracts video frames with ffmpeg and sends them to Claude as base64 images with timestamps, alongside an audio transcription. Audio comes from the Gemini API, local Whisper (whisper.cpp or Python openai-whisper), or the OpenAI Whisper API. It also accepts YouTube URLs via yt-dlp, using manual or automatic captions before falling back to the audio backend.

Who it is for: Claude Code users who want Claude to analyze local or YouTube videos

Strengths

  • Three audio backends, including fully offline whisper.cpp with automatic model download
  • YouTube URLs use manual or auto captions first, with transcription_source labeled
  • Six MCP tools, including video_analyze and video_detail for drilling into specific moments
  • Frame fps, time range, resolution and format (jpeg, png, webp) are adjustable

Weaknesses

  • Requires Node.js 24+ and ffmpeg installed locally
  • YouTube support needs yt-dlp installed separately
  • Only a perception layer: no interpretation, and frames count against Claude's context
  • Windows users may need a global npm install workaround if npx fails

What it needs

  • GPU optional
  • Needs Node.js 24+, ffmpeg, yt-dlp (optional, YouTube), whisper.cpp or openai-whisper (local backend), Gemini API key or OpenAI API key (cloud backends)
  • Models: Gemini API, whisper.cpp, openai-whisper, OpenAI Whisper API

Quick answers

Is claude-video-vision open source?

Yes. Its code is on GitHub under the MIT license.

Also in Skills, plugins and prompts

See all 324
Also in Skills, plugins and prompts
RankProjectScore
1impeccableDesign skill with 24 slash commands for AI coding agents' frontend output Live demo ↗ (opens in a new tab)98.9k starsApache-2.096 out of 100
2skillsAgent skills for planning, TDD, debugging and code review308.9k starsMIT95 out of 100
3eccAgents, skills, hooks and rules bundle for Claude Code, Codex and other harnesses294.5k starsMIT95 out of 100
#4gstackMarkdown slash-command skills that give Claude Code planning, review, QA and ship roles138.5k starsMIT95 out of 100
#5cavemanCuts token use in coding agents with terse replies and compressed inputs Live demo ↗ (opens in a new tab)116.6k starsApache-2.095 out of 100