A raw caption track is nearly unreadable and expensive to feed into a model. Rendering it as Markdown — with chapters if the video has them, or with only the matching passages when you supply a search term — is the difference between quoting a video and pasting it. The search template is the one to reach for on a long talk.
A server for reading videos rather than watching them: it fetches metadata and caption tracks — auto-generated or manual — and renders them through templates so the transcript arrives as a document rather than a wall of timed lines.
- `get_video_info` returns the video's metadata from a URL
- `get_captions` pulls the caption track, with a `language` parameter that defaults to English; English and French are supported out of the box
- `convert_to_markdown` renders the transcript through one of three templates — basic for a plain transcript, detailed for full metadata with timestamps, search for matches highlighted in context — with options for chapters and a search term
- `list_templates` shows what is available, and the template list is editable in the source
Node.js v16 or higher. Clone the repository, install, build, and point the client at the compiled entry point; published as youtube-mcp, version 0.1.0. A YouTube Data API key in YOUTUBE_API_KEY covers public videos. Private videos need OAuth2 — a project in Google Cloud Console with the YouTube Data API v3 enabled, then YOUTUBE_CLIENT_ID, YOUTUBE_CLIENT_SECRET and a refresh token obtained through the included script.
Build from source — clone the repository and build it, then point your client at the binary