RedNote transcript: get the transcript of a RedNote (Xiaohongshu) video from its link

Short answer: RedNote (小红书, Xiaohongshu) does not publish captions or a transcript for its videos, and ChatGPT cannot open a RedNote link at all. To get a transcript, paste the note's share link into LinkDigest: it reads the public share page, transcribes the speech into timed segments, reads the text burned into the frames, and returns it as JSON or Markdown. No RedNote account, no cookies. Sign-up comes with 10 free credits.

Disclosure: LinkDigest is the author's own service.

Why there is no "transcript" button

RedNote's web player has no caption track to download. The subtitles you see in many videos are burned into the picture, so they are pixels, not text. A transcript has to come from two places: the audio (speech recognition) and the frames (reading the on-screen text). LinkDigest does both and keeps them apart, so you can tell what was said from what was written on screen.

The link you need

Copy the link from the RedNote app with Share → Copy link. The link must carry its xsec_token — RedNote only serves a note to a link that has one; a bare /explore/<id> address usually returns an empty page. You can paste the whole share text the app copies; the link is picked out of it.

One call

curl -X POST https://linkdigest.dev/api/v1/digest \
  -H "Authorization: Bearer ld_live_..." \
  -H "Content-Type: application/json" \
  -d '{"url": "https://www.xiaohongshu.com/discovery/item/...?xsec_token=...", "depth": "transcript", "format": "json"}'

depth: "transcript" returns the speech and the on-screen text and nothing else, at 1 credit per started two minutes of video. Leave depth out to also get the caption, the images described, the top comments and the key points, each backed by a quote checked against the note.

What comes back

Field	What it holds
`transcript[]`	`{t, text}` segments, `t` in seconds from the start
`on_screen[]`	Text read from the frames, with the time it appears
`title`, `caption`, `author`, `posted_at`	The note as posted
`stats`	Likes, collects, comments as of the read
`coverage`	How many seconds were transcribed, so a partial read is never silent

Ask for format: "markdown" and the same arrives as a page you can paste into a document or a prompt.

From an agent

The same read is a tool in Claude Code, Cursor and any MCP client: add https://linkdigest.dev/mcp with your key and ask for the transcript of the link. The Xiaohongshu platform page has the setup for each client, and the API docs the full response. For image notes — most of RedNote — see the Xiaohongshu (RedNote) API post; for Douyin, the Douyin transcript post.

Limits, said plainly

A note whose link has no xsec_token, or that its author deleted, returns a clear error and costs nothing.
Speech recognition is Mandarin-first; heavy dialect or music over speech lowers accuracy, and coverage tells you how much was transcribed.
Nothing is posted, liked or followed; LinkDigest reads the public page and stops there.

This is the write-up behind LinkDigest, which does the above as a hosted service — with an MCP server for Claude Code and Cursor, a REST API, and 10 free credits to try it.