The social media reading layer for AI agents
One call turns a Xiaohongshu, Douyin, YouTube, TikTok or X link into text: transcript, on-screen text, image descriptions and metadata — read by us, so no cookie or account of yours is ever involved. Add breakdown for the hook, the beats and a reusable template. Markdown for reading, JSON for pipelines. $0.02 an image post, $0.02 per minute of video, 10 free on sign-up.
Reading the post…
why linkdigest
Xiaohongshu puts whole articles inside pictures; Douyin and TikTok say it out loud. We OCR every image, transcribe the speech with timecodes, and read on-screen text with when it appeared.
One link: a plain fetch got 860 characters. LinkDigest returned 35,059.
No cookies to export, no login to keep alive, no account of yours at risk of a ban. We do the reading; you send a link.
Works from a server, an agent or a script — nothing to install but one MCP line.
Tick viral breakdown and get the hook, the structure as timed beats, the title formula, the call to action and a reusable template.
Every quote is checked word for word against the post; engagement and hashtags come from the platform, not the model.
Claude Code and Cursor over MCP, a REST call, batches of 50 with a signed webhook, Dify and Apify — and an OpenAPI spec (/openapi.json) for Coze and other tool builders.
10 free credits to start. Links anyone has already read cost nothing, and a read that finds nothing is not charged.
try it
One free digest, no account. The examples below run instantly because someone has already digested them — cached links are always free.
Try an example below, or paste your own link.
real output
Four public digests, one per platform. Every line came out of the engine.
the problem
Both sides below are real output for the same Xiaohongshu post. On the left, what your agent gets today. On the right, what it gets through LinkDigest.
platformxiaohongshu
author周星星
title从 Agent Team 到 Agent Swarm（下篇）
images · 17 described and read
ocr_text · 381 fragments
key_points · 13
how it works
Short links are expanded and tracking parameters stripped, so two people sharing the same post hit one cached result instead of paying twice.
Captions are used where a platform publishes them. Otherwise the audio is transcribed and frames are sampled for burned-in text — never the whole file, and never stored.
One structured object: transcript with timecodes, on-screen text, image descriptions, key points, and the whole thing pre-rendered as Markdown.
coverage
Fetching other people’s platforms breaks, so this table is kept honest rather than aspirational. When a platform blocks us, LinkDigest says so instead of returning a confident description of an error page.
| Platform | State | Notes |
|---|---|---|
| xiaohongshu 小红书 | live | Image notes (up to 60 images) and video notes. No login needed. |
| douyin 抖音 | live | Video posts and image notes, including 图文. Long videos can be read transcript-first. |
| youtube | live | Read by Gemini watching the video, since YouTube blocks datacenter IPs. |
| tiktok | live | Short links resolve. Speech is read from the video itself, not the sound track. |
| x | live | Video posts in full. Image posts: every photo described. Long-form articles: the lead image. |
| wechat 公众号 | live | Public-account articles (mp.weixin.qq.com). No login needed. |
| web pages | live | Article text and metadata. |
| bilibili | blocked | Bilibili refuses our server's address (HTTP 412). Needs a proxy. |
| soon | Wired, not yet verified end to end. |
what you don’t have to do
Each of these is something that was built, checked against a real link, and is written down in the coverage table or the docs.
Image notes and video notes are read with no account at all, so there is nothing of yours to get banned.
Douyin refuses anonymous requests, so it runs on our session cookies. When they lapse, that is our pager, not your incident.
YouTube blocks datacenter address ranges, so a server-side fetch fails wherever it is deployed. Here a model watches the video instead.
Every image in a Xiaohongshu note is described and read for on-screen text — the 17-image example on the homepage came back with all 17.
A digest that is thinner than the post says so in plain words, in a field your code can branch on, instead of returning a confident description of less.
If no transcript, text or image came back, the digest is not charged and is retried within the hour instead of being cached for a month.
Anything anyone has already digested comes back instantly and never counts against you.
install
The agent calls digest_url by itself when it meets a social link. You don’t have to ask.
MCP · Claude Code
claude mcp add --transport http linkdigest \ https://linkdigest.dev/mcp \ --header "Authorization: Bearer ld_live_..."
REST
curl -X POST https://linkdigest.dev/api/v1/digest \
-H "Authorization: Bearer ld_live_..." \
-d '{"url": "https://v.douyin.com/..."}'before you trust it
No company behind this, so here is the evidence instead of a logo wall.
116 reads by users, 58% came back whole; a whole read took 27 s at the median. Failures on our side count against us; our own accounts are left out; per-platform figures are recomputed hourly.
Published as dev.linkdigest/linkdigest. The registry only grants a domain namespace to whoever controls the domain's DNS, and the proof is public.
The SDK, CLI and skills are MIT on GitHub. Releases are built by GitHub Actions and carry a build attestation you can verify with `gh attestation verify`.
A call is a link, nothing else. No cookies, no passwords, no phone number; the reading happens on our servers, so nothing can get your account flagged.
Media is read and discarded. We never store, serve or redistribute the video or images behind a link; comments and private posts are not read.
Cards and Alipay are handled by Stripe. Card numbers never touch our servers, and the 250-credit pack is a one-off, not a subscription.
Run by one developer in Malaysia. Mail support@linkdigest.dev and a person replies, usually the same day. Domain ownership proof is published too.
pricing
Digests are cached and shared, so a link someone already read comes back instantly and never counts against you.
$0
10 credits when you sign up — once, they do not renew. Every platform, full output, no card. Then $5 once for 250 credits, no subscription.
$9/month
500 credits a month — about 500 minutes of video, or that many posts.
Every plan includes the REST API and the MCP server — the free plan too.
questions