Extract transcripts and subtitles from YouTube videos, playlists and whole channels. You are only charged for transcripts that actually arrive.

Most transcript scrapers talk to YouTube's WEB player client. In 2026 that client answers caption requests with UNPLAYABLE, or returns a caption URL serving an empty body with 200 OK. That is why scrapers fail partway through a run and bill you anyway.

This one:

• Falls back across player clients (ANDROID, IOS, TVHTML5) until one returns real caption data, then verifies it contains actual text.
• Retries on a fresh IP. YouTube throttles by address, so a failed lookup is repeated from a different residential exit.
• Runs a recovery pass. Anything still failing gets a second, slower sweep, so a video is only called unavailable after failing a quiet retry.

Measured: a 40-video run across two channels returned 40 of 40. Without the residential default, the same run returned 4 errors out of 5.

Put video links, playlist links, channel links, @handles or bare video IDs into videoUrls. Playlists and channels expand automatically. Everything else is optional: preferred languages, translateTo, outputFormat (segments, text, srt, vtt), maxVideos, concurrency.

One record per video: title, channel, duration, view count, available languages, whether captions were human-written or auto-generated, a timestamped transcript, and a flat text field ready for RAG.

srt and vtt output trims overlapping cues, so subtitle files are usable as-is.

Videos that cannot be scraped return a plain error and are not billed.

• Videos with no captions cannot be transcribed. This reads YouTube's caption tracks, it does not run speech-to-text.
• Private, members-only and age-restricted videos are not accessible.
• Live streams have no complete transcript until they finish.