The quickest way, with no setup
YouTube generates transcripts for most videos automatically. Open the video, expand the description, and use "Show transcript" — you can select the text and paste it into Claude directly. For a single video this beats installing anything.
Two caveats. Auto-generated transcripts have no punctuation and mis-hear proper nouns and jargon, so a technical talk may arrive noticeably garbled. And a long video produces a lot of text: an hour of speech is roughly 9,000 words, which is fine for Claude's context window but will crowd out other material in a long conversation.
Pasting also loses timestamps unless you keep them, which matters if you want to ask "where in the video does she talk about X" rather than just summarizing.
Using an MCP server instead
If you do this regularly, an MCP server lets you give Claude a URL and have it fetch the transcript itself. The practical difference is not effort per video so much as what becomes possible: Claude can pull several videos in one conversation, or fetch a transcript midway through a task without you leaving the chat.
Transcript servers are generally read-only and need no API key, which makes them one of the lower-risk things to connect. They work by requesting the caption track YouTube already publishes, so they are subject to the same limitation as the manual approach — a video with captions disabled has nothing to fetch, and no server can work around that.
Search our catalog for YouTube transcript servers and check the detail page before installing. The maintenance signals matter here more than usual: this category depends on an undocumented YouTube endpoint, so a server that stopped being updated is likely to be broken rather than merely stale.
When the video has no captions
Some videos genuinely have no caption track — the uploader disabled them, or the video is too new for YouTube's automatic pass. In that case there is no transcript to fetch by any method, and the remaining option is to transcribe the audio yourself with a speech-to-text tool, then paste the result.
This is worth knowing before you start debugging: a transcript server returning nothing for one specific video is usually correct behavior, not a bug.
Which approach to pick
For one video, paste it. Setting up an MCP server to summarize a single talk is strictly more work than selecting text.
For repeated research across many videos, install a server. The break-even is somewhere around the third or fourth video in a session.
For anything where accuracy of specific quotes matters, verify against the video itself regardless of method. Auto-generated captions are good enough to summarize from and not good enough to quote from.