agents-libagents-lib
AgentsMCP ServersWorkflowsTutorialsComparisonsTemplatesNewsAbout
🔌

Claude Video

Multimedia•⭐ 13,000

Skill that gives Claude and 50+ AI agents the ability to watch and analyze any video.

GitHubWebsite →
Claude CodeCodexCursorCopilotGemini CLI

Overview

Skill that gives Claude and 50+ AI agents the ability to watch and analyze any video.

Setup

Run with npx:

npx skills add bradautomates/claude-video -g

Configuration

Auto-installs ffmpeg and yt-dlp on macOS; Whisper API optional

Documentation

Claude Video

Overview

Claude Video is an open-source tool that gives Claude (and other AI assistants) the ability to watch and analyze any video — downloading it, extracting frames, transcribing audio, and passing everything to the AI for grounded answers. At 13,000+ GitHub stars, it has become the go-to skill for video intelligence in the AI agent ecosystem.

Built by Brad Bonanno (Solaris Automation), Claude Video works with Claude Code, Codex, Cursor, Copilot, Gemini CLI, claude.ai (web), and 50+ other agent hosts — making it one of the most broadly compatible AI skills available.

Features

  • Multi-surface support: Claude Code, Codex, Cursor, Copilot, Gemini CLI, claude.ai, and 50+ agents
  • Any video source: URLs via yt-dlp (YouTube, TikTok, X, Instagram, Loom, and hundreds more) or local files (.mp4, .mov, .mkv, .webm)
  • Four detail modes:
    • transcript — captions only (fastest)
    • efficient — keyframe-based
    • balanced — scene-aware, capped at 100 frames
    • token-burner — uncapped full analysis
  • Free captions first: Uses native captions before falling back to Whisper
  • Frame deduplication: Drops near-identical frames to save tokens
  • Focused analysis: --start and --end flags for segment-level zoom
  • Zero config: Auto-installs ffmpeg and yt-dlp on macOS; prints setup commands on Linux/Windows
  • MIT license

Installation

Claude Code:

/plugin marketplace add bradautomates/claude-video
/plugin install watch@claude-video

Other agents:

npx skills add bradautomates/claude-video -g

claude.ai (web):

  1. Download the .skill bundle from the GitHub releases page
  2. Go to Settings → Capabilities → Skills
  3. Upload the .skill file

Usage

Basic video analysis:

> /watch https://www.youtube.com/watch?v=abc123

What are the key points of this video?

# Claude Video:
# - Downloads video via yt-dlp
# - Extracts frames at balanced detail level
# - Pulls transcript from captions or Whisper
# - Feeds frames + transcript to Claude
# Claude responds with a grounded summary

Segmented analysis:

> /watch https://www.youtube.com/watch?v=abc123 --start 2:30 --end 5:00

Explain what happens between 2:30 and 5:00 in detail.

Detail mode control:

# Fast transcript-only analysis
> /watch --detail transcript https://example.com/video.mp4

# Maximum detail (highest token cost)
> /watch --detail token-burner https://youtube.com/watch?v=abc

How It Works

  1. User pastes a URL or local path with a question
  2. yt-dlp checks for captions; downloads only what's needed
  3. ffmpeg extracts frames at the chosen detail level
  4. Transcript is pulled from native captions or Whisper API (Groq/OpenAI)
  5. Claude reads every frame as an image alongside the timestamped transcript
  6. Answers are grounded in what's actually on screen and in the audio

Common Use Cases

  • Competitive analysis: Watch and analyze competitors' product demos and launches
  • Bug diagnosis: Review screen recordings of bugs with full context
  • Long-video summarization: Distill hour-long content into concise notes
  • Changelog distillation: Extract actual changes from verbose update videos
  • Playlist-to-notes: Convert entire video playlists into searchable knowledge
  • Meeting recaps: Analyze recorded meetings and extract action items

Pros

  • ✅ Works with 50+ agent hosts — unmatched compatibility
  • ✅ Handles any video source (YouTube, TikTok, X, local files)
  • ✅ Four detail modes balance cost vs. thoroughness
  • ✅ Free captions first — no API costs when captions exist
  • ✅ Frame deduplication reduces token waste
  • ✅ Segment-focused analysis for dense content
  • ✅ Zero-config setup on macOS
  • ✅ MIT license

Cons

  • ❌ token-burner mode is expensive for long videos
  • ❌ Whisper fallback requires Groq or OpenAI API key
  • ❌ Frame extraction limited by system resources
  • ❌ yt-dlp compatibility varies across video platforms

When to Use

  • Content analysis — understanding video content for research or summary
  • QA/testing — diagnosing bugs from screen recordings
  • Competitive intelligence — monitoring product launches and demos
  • Education — extracting lessons from long-form video content
  • Meeting notes — automated recaps from recorded sessions

Resources