claude-video-vision

Why we included this project

This is a Claude Code plugin that gives Claude a way to actually see and hear video. Rather than building its own vision pipeline, it acts as a perception layer: ffmpeg pulls frames out of a local file or YouTube URL, an audio backend (Gemini, local Whisper, or OpenAI) returns a timestamped transcript, and Claude reads both as evidence. That split works well for real jobs like debugging a screen recording, summarizing a long lecture, or checking what text shows up at a specific moment in a clip. It installs from a marketplace command with no build step, Claude tunes the extraction to how you phrase the question, and cloud or fully local backends both work, so transcription can stay on your machine if you prefer. The plugin intentionally leaves the reasoning to the underlying model, feeding Claude evidence rather than conclusions.