YouTube Transcript Server
YouTube Transcript Server is an MCP server that retrieves captions and subtitles from YouTube videos directly, with optional timestamps, automatic language fallback, and built-in filtering of ads and sponsorship segments.
This Model Context Protocol server gives an AI agent direct access to a YouTube video's existing captions and subtitles through one tool: it accepts a video URL, a Shorts URL or a video ID, an optional language code with automatic fallback to an available language, an option to include timestamps, and an option, on by default, to strip sponsorship and promotional content based on chapter markers.
A second, optional tool analyzes a video with the TwelveLabs Pegasus video-understanding model instead of reading its captions, so it can summarize or answer questions about what is shown on screen even for videos with little or no speech, such as demos, gameplay or b-roll; it requires a direct video file URL, since it cannot fetch a YouTube watch page itself, and needs a separate TwelveLabs API key.
It is written in TypeScript and can be installed for Claude Desktop through Smithery.