I built mcptube because I spend a lot of time learning from Stanford and Berkeley lectures on YouTube, and I wanted a better way to interact with the content.

mcptube ingests YouTube videos (transcripts, metadata, frames) and exposes them through both a CLI and an MCP server. You can semantically search transcripts, extract frames at any moment, ask questions about videos, and generate illustrated reports.

It works with Claude Code, VS Code Copilot, Claude Desktop, and any MCP-compatible client. The MCP tools are passthrough — the connected LLM does the heavy lifting, zero API key needed on the server. The CLI also supports BYOK for more deterministic results.

Tech stack: FastMCP, yt-dlp, ChromaDB, LiteLLM, SQLite.

GitHub: https://github.com/0xchamin/mcptube PyPI: https://pypi.org/project/mcptube/

pipx install mcptube --python python3.12