claude-real-video
Let Claude (or any LLM) actually watch a video — scene-aware, deduplicated frames + transcript, from a URL or local file. Runs locally, MIT.
claude-real-video is a local video analysis tool that enables any LLM—including Claude, ChatGPT, and Gemini—to understand video content through intelligent scene detection and transcription. It extracts only the frames that matter (scene changes, not fixed intervals), removes near-duplicates using sliding-window deduplication, and transcribes audio with Whisper, all processed locally on your machine. Available as a Claude Code plugin, MCP server, agent skill, or standalone CLI, it converts YouTube URLs or local video files into timestamped keyframes and transcripts that models can cite. Unlike cloud services that sample at fixed intervals or only read video transcripts, claude-real-video performs scene-aware extraction so fast cuts aren't missed and static content doesn't flood the context window.
Key Features
Use Cases
- 01Enable Claude Code, Cursor, or other coding assistants to analyze screen recordings, tutorials, or demo videos during development workflows
- 02Extract and cite specific visual evidence from video interviews, meetings, or presentations with frame-level timestamps
- 03Process educational content (lectures, coding tutorials) where both slides and spoken explanations need to be captured together
- 04Analyze marketing videos, product demos, or competitor content to extract messaging, pricing strategies, or feature comparisons
- 05Create searchable archives of video libraries where you can query across transcripts and visual content from multiple videos
- 06Process login-gated or locally-stored video content without uploading to cloud services, maintaining privacy and data control
Related Skills
View moresuperpowers
An agentic skills framework & software development methodology that works.
skills
Skills for Real Engineers. Straight from my .agents directory.
skills
Public repository for Agent Skills
ponytail
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
claude-real-video — FAQ
What is claude-real-video and how does it help AI assistants watch videos?+
claude-real-video is an open-source Python tool that converts videos into AI-readable formats by extracting scene-aware keyframes and generating timestamped transcripts. Unlike fixed-interval sampling that misses fast cuts or floods context with duplicates, it uses intelligent scene detection and deduplication, processing everything locally so you control what gets shared with LLMs.
How do I install claude-real-video for Claude Code or other AI assistants?+
First install ffmpeg (brew install ffmpeg on macOS), then run pip install "claude-real-video[whisper]" followed by npx skills add HUANGCHIHHUNGLeo/claude-real-video to add it as an agent skill. For Claude Code plugin marketplace, use /plugin marketplace add HUANGCHIHHUNGLeo/claude-real-video then /plugin install claude-real-video@claude-real-video.
Which AI clients and assistants work with claude-real-video?+
It works as an agent skill with Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI, and 50+ agentskills.io-compatible hosts. It also functions as an MCP server for Claude Desktop and Cursor, or as a standalone CLI (crv) whose output can be pasted into any LLM including ChatGPT and Gemini.
Does claude-real-video require API keys or upload videos to the cloud?+
No API keys are needed—all processing (frame extraction, deduplication, Whisper transcription) runs locally on your machine. Videos are never uploaded; only the extracted frames and transcript you choose to paste into an LLM afterward are sent to that provider.
Is claude-real-video free and open source?+
Yes, the core tool is MIT-licensed and completely free. A paid add-on (crv Pro, $29 one-time) adds camera motion analysis, voice emotion detection, and interactive breakdowns, but the base version handles all keyframe extraction and transcription at no cost.
What are the system requirements and dependencies?+
Requires Python 3.10+ and ffmpeg/ffprobe (installed via brew, apt, or winget depending on your OS). The [whisper] extra installs OpenAI Whisper for transcription; optional [fast], [mlx], and [speakers] extras enable faster processing, GPU acceleration on Apple Silicon, and speaker diarization respectively.
How do I install claude-real-video?+
Open the source repository on GitHub and follow its README. claude-real-video is a skill — MCP Agents Market links you directly to the official repo.
Is claude-real-video free?+
claude-real-video is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.