I built a small browser extension that exports the transcript of the YouTube video you are watching — as TXT, Markdown, SRT, or VTT, in one click from the toolbar popup. This post is a plain account of what it does, how it gets the transcript, and where it stands.
What it does
Open a YouTube video, open the extension popup, and export the transcript. The free version exports a single video in four formats:
- TXT — plain text, one cue per line
- Markdown — timestamped, readable in any notes app
- SRT — standard subtitle file
- VTT — WebVTT subtitle file
Two paid features are planned, not for sale yet: bulk export of a whole playlist, and bilingual side-by-side transcripts. The extension is free to install; there is no account and no server.
How it gets the transcript
There are two paths, and the extension tries them in order:
- The timedtext endpoint. This is the full transcript, straight from YouTube's own caption data. When it works, you get every cue.
- The transcript panel on the page. If you open the transcript panel on the watch page (the "..." menu under the video, then "Show transcript"), the extension reads the rendered rows as a fallback.
I tested both paths against a real video (dQw4w9WgXcQ) in a real browser. The panel path returned 24 cues and 2,066 characters, and all four formats were built from those real cues: TXT 2,335 bytes, Markdown 2,668 bytes, SRT 3,142 bytes, VTT 3,162 bytes. I spot-checked the content; the timestamps and text matched the video.
Two things I measured that changed the design
YouTube changed the transcript DOM. The old selector, ytd-transcript-segment-renderer, now returns zero rows. The current rendering uses <transcript-segment-view-model>, with the timestamp in .ytwTranscriptSegmentViewModelTimestamp and the text in span[role="text"]. The extension tries the new DOM first and falls back to the old one.
The timedtext endpoint returns empty bodies from datacenter IPs. I measured this twice: from curl (all formats HTTP 200 with 0-byte bodies, then 429 after repeated requests) and inside a real browser (json3 through srv3 all empty, no challenge page). The panel fallback is what works in that environment. On a normal residential connection the timedtext path is expected to work and give the full transcript — but I have not been able to verify that from this environment, so I am stating it as an expectation, not a fact.
The panel only gives the rows YouTube has rendered so far (24 on the first screen in my test), so it is a partial transcript. The timedtext path remains the only full-transcript route.
Status
Version 0.1.0 is submitted to Microsoft Edge Add-ons and is in review since 2026-11-15. It is not installable yet. The other extensions I am building the same way — FocusBlock (live on Edge), ConvoExport and SessionSnap (both in review) — follow the same pattern: local-first, no account, no tracking.
The extension's test suite is 120 tests, all passing. Its privacy page is here: https://alapha888.github.io/yt-transcript-export-privacy/
Why this exists
There are good transcript tools already — Transcript for YouTube has around 40,000 users, and Glasp is far larger with a paid Pro tier. What I did not find is a small, local-first exporter that treats the transcript as a file you own: no account, no upload, four plain-text formats, and a playlist bulk export for later. That is the whole pitch. If the review comes back, I will update this post with the store link.
Top comments (0)