I built a tool that gives Claude Code eyes on videos — extract code from any tutorial

I was tired of pausing YouTube tutorials and manually typing code off the screen. So I built videocode.

What it does

Point it at any coding video → get working, runnable code out.

• 🔍 Extracts key frames with scene detection
• 🎙️ Transcribes audio with Whisper for context
• 🤖 Uses Ollama vision models (llava, llama3.2-vision) to read code off screen
• 📁 Assembles a full project structure

Quick start

\bash
pip install videocode
ollama pull llama3.2-vision
videocode code "https://youtube.com/watch?v=..."
\\

MCP server for Claude Code

Add to your ~/.claude/settings.json:

\json
{
"mcpServers": {
"videocode": {
"command": "videocode",
"args": ["mcp"]
}
}
}
\\

Then just ask Claude Code: "Extract the code from this tutorial: https://youtube.com/..."

Backends

Backend	Cost	Privacy
Ollama (llava)	Free	100% local
Ollama (llama3.2-vision)	Free	100% local
Gemini 2.5 Flash	Free tier	Cloud
OpenAI GPT-4o	Paid	Cloud

100% local by default. No API key required.

GitHub: https://github.com/marianoperezbaldasare-maker/videocode