get_transcript takes url and language and returns the transcript; there is no search tool, no metadata call, and no way to ask in advance which languages a given video actually carries. The workflow this fits is one where the video was already chosen somewhere else and the language is known — it will not help you find either. That narrowness is the whole design, and it is worth knowing before it is wired into a pipeline that assumes discovery.
A single-tool server that returns the transcript of a YouTube video from its URL.
- get_transcript takes the video url and a language, and returns the transcript text.
- That call is the whole surface: there is no video search and no separate listing of which languages a video carries, so language is the only choice being made at call time.
Nothing to supply: no account, no key. get_transcript needs the URL of a video that carries a transcript, and the language to pull it in.
One command — npx mcp-remote https://youtube-transcript-mcp.ergut.workers.dev/sse --debug