Short-form video intelligence API
Give your assistant something to call when someone asks what is working on TikTok.
ShortsIntel is a headless API and a hosted MCP server for TikTok, Instagram Reels and YouTube Shorts. Search, per-video Intelligence, creator baselines and Scores, daily Trackers with Change Reports, and a Research operation that turns an intent into a Digest — all in compact JSON built for model context rather than raw vendor payloads.
Prepaid, priced per operation. New accounts start with $5 of credit. Agents can read /llms.txt and /llms-full.txt directly.
The gap this fills
People who make or market short-form video increasingly work through AI assistants. When they ask “what is working in my niche”, “is this creator’s last video an outlier”, or “watch this hashtag and tell me when it moves”, the assistant has nothing to call.
Scraping vendors return large per-platform payloads, keep no history, and offer no judgement. So the assistant burns tokens re-deriving the same facts every time, and still cannot answer any question about change over time.
ShortsIntel sits on top of that raw data and does the parts an agent should not have to: normalising three platforms into one shape, running the video pass that turns a clip into structured Intelligence, and keeping a shared Corpus of Snapshots so yesterday is still there to compare against.
That history is what makes an outlier multiple or a velocity number possible at all — and it compounds, because every lookup anyone makes writes a Snapshot the next question can use.
One envelope, one price, no surprises
Every operation is a Run
A call holds the response open for up to {wait} seconds. If the work finishes, you get the result inline. If it does not, you get a run id to poll or receive by webhook. Same shape either way, so you handle one response type.
Ask the price first
estimate: true on any paid operation returns what it would cost and whether the answer is already cached — without creating a Run or charging anything. An agent can ask before it spends.
You pay for delivered work
Cached results cost 25% of fresh. Failed runs charge nothing. A partial run keeps what completed and charges per item. Re-reading a finished run is free forever, so you never pay twice for the same result.
It says what it does not know
Instagram’s index is best-effort, so results carry coverage: "partial" rather than quietly returning less. Errors are problem+json with a stable code to branch on.
A search response:
{
"run": {
"id": "run_2f8c…",
"kind": "VIDEO_SEARCH",
"status": "completed",
"charge_usd": 0.01,
"cached": false
},
"result": {
"items": [ … ],
"next_cursor": "eyJ0…",
"coverage": { "tiktok": "complete", "instagram": "partial" }
}
}What you can call
Research
Give it an intent in plain language. It derives keywords, searches every platform, de-duplicates, enriches the top videos, ranks them and returns a Digest: themes with shares, the hooks that work quoted verbatim, sentiment, and recommended angles — plus the ranked videos, creators, hashtags and sounds behind it.
$0.50 for up to 50 enriched videos
Video Intelligence
One structured object per video: hook and opening line, call to action, sponsorship signals, brands, trend references, speaking style, visual format, on-screen text, setting, camera and lighting, and a brand-safety tier with reasons. Videos with no speech still run.
$0.02 per video, $0.005 when re-read from cache
Scores
Every creator gets a baseline from their last 20 videos: median views, p90, engagement rate, posting cadence, sample size and confidence. A video is then an outlier multiple, a views-per-hour velocity and a percentile against that baseline. Numbers, not labels — you decide what counts as a hit.
$0.05 per score
Trackers and Change Reports
Watch a creator daily. Snapshots accumulate and are kept forever; a Change Report arrives only when something actually moved — new videos, an outlier, followers ±10%, a velocity spike — so your agent is not woken by a quiet day.
$0.05 per cycle, +$0.02 per new video with intelligence
Search and lookups
Keyword search across TikTok, Instagram Reels and YouTube Shorts, normalised into one shape. Plus video, creator, hashtag, sound, ad-library and comment lookups, priced near cost because they are the primitives everything else builds on.
From $0.005 per lookup
Webhooks
Signed callbacks on run.completed and tracker.changed, with an HMAC signature, a delivery log, three attempts over about an hour, replay for the ones that never landed, and secret rotation with a 24-hour overlap so a deploy never drops an event.
Free
Full catalog, parameters and prices in the API reference.
Connect it in a minute
The same 16 operations are available as MCP tools or as REST endpoints. Claude.ai connects over OAuth with no key at all; Claude Code, Cursor and VS Code take a bearer key.
Claude Code, one command:
claude mcp add --transport http \
shortsintel https://www.shortsintel.com/api/mcp \
--header "Authorization: Bearer sk_live_…"Or straight over HTTP:
curl -X POST https://www.shortsintel.com/v1/research \
-H "Authorization: Bearer sk_live_…" \
-H "Content-Type: application/json" \
-d '{"intent":"what is working in cold plunge"}'What we hold, and what you may do with it
- ShortsIntel processes public short-form video data — posts, captions, public metrics, public profiles — under legitimate interest.
- Media files are deleted after the Intelligence pass. Raw comments are purged after 90 days; the sentiment analysis is kept and never carries usernames.
- Snapshots are shared across customers and kept indefinitely, which is what makes history available from your first call.
- Derived works are allowed: the analyses, reports, scores and dashboards you build on the output are yours to sell.
- Reselling is not: republishing bulk raw records, mirroring the corpus, or using the data to target or profile individuals are prohibited.
- Read the terms and the privacy policy.