Claude Video Vision
Lets Claude 'watch' a video: extracts frames via ffmpeg and transcribes audio (Gemini, local Whisper or OpenAI) with timestamps.
How to set it up
- Install Node 24+ and ffmpeg.
- Install the plugin; the MCP server auto-installs via npx.
- Set GEMINI_API_KEY or install whisper-cpp.
- Run /watch-video on a file.
This is a plain-English summary of the project's own instructions. Commands and versions change, so always follow the current README at the source link.
Safety notes
- Gemini or OpenAI backends send audio to that provider and need an API key; local Whisper keeps it on-device.
General rule: open a duplicate of your project before letting any agent make changes, and read what a script does before approving it.
Alternatives and related tools
HyperFrames (HeyGen)
Open-source framework that turns HTML, CSS, media and seekable animations into deterministic MP4s; usable from AI coding agents through skills.
ffmpeg-skill
42 local FFmpeg tools (captions with karaoke animation, 9:16 fit, silence removal, straighten, etc.) with a machine-readable contract; the agent checks its own output.
FableCut
Zero-dependency browser video editor with a JSON timeline that AI agents drive via MCP or REST, with a live-reloading UI.
Kinocut (formerly mcp-video)
Guardrailed editing MCP: local FFmpeg tools, 'Video Receipts', quality gates and Shorts/Reels repurposing, with 200+ MCP tools.
VibeFrame
Turns a brief into a plan, generates assets with Seedance, Runway, Kling or Gemini on your own keys, and renders an MP4 under a hard spending cap.
FFmpeg-MCP (video-creator)
Find local videos, trim, stitch and play them through chat using the ffmpeg command line.
The AgentCut weekly
New AI connectors for After Effects, Premiere, Resolve and CapCut, tested setup notes, and what broke after app updates. One email a week.