GET /v1/transcript

Any video URL, returned as clean, usable text.

One endpoint turns TikToks, tweets, talks and podcasts into transcripts your pipeline can use directly. When captions don't exist, speech recognition takes over — any video with a human voice returns text.

Start free — 25 calls a month Connect an agent via MCP

median short-form turnaround 7 s^† · no card required

Transcribing tiktok.com/@garyvee

Request

curl "https://api.payload.cool/v1/transcript
      ?url=https://www.tiktok.com/@garyvee/video/76720…"
  -H "X-API-Key: YOUR_KEY"

import requests

r = requests.get("https://api.payload.cool/v1/transcript",
      params={"url": video_url},
      headers={"X-API-Key": KEY})
text = r.json()["transcript"]

const r = await fetch(
  `https://api.payload.cool/v1/transcript?url=${url}`,
  { headers: { "X-API-Key": process.env.KEY } });
const { transcript } = await r.json();

Response · 2008 901 ms

{
  "platform":   "tiktok",
  "title":      "YOU HAVE TIME",
  "author":     "garyvee",
  "duration_s": 46.0,
  "route":      "asr",  ← no captions; audio transcribed
  "transcript": "I'm 62 and I did all the right things
                 as a father, and last year I was able
                 to leave and just… live my life."
}

Caption scrapers return nothing when captions don't exist

Most transcript APIs read the platform's subtitle track and give up when it's missing — which, on TikTok, is most of the time. This one transcribes the actual audio.

	Caption scrapers	Video Transcript API
Video without captions	empty response	full transcript, via speech recognition
TikTok photo posts	error	on-image text returned
X, Vimeo, Twitch, TED, podcasts	one API per platform	the same endpoint
AI-agent integration	build it yourself	hosted MCP server
When things fail	silent junk	named error codes, honestly documented

One endpoint, the whole video web

Purpose-built extraction for TikTok; a battle-tested fallback engine for everything else. Where support is honest-but-partial, it says so.

TikTokvideos & photo posts, no loginFULL

X / Twitternative video postsFALLBACK

Vimeocaptions preferred when presentFALLBACK

TwitchVODs and clipsFALLBACK

TEDa 19-minute talk in 70 secondsFALLBACK

Podcast hostsdirect episode URLsFALLBACK

1,800+ othersanything the extractor can reachFALLBACK

YouTubea clear 422 until it meets the bar — never a silent failureSOON

Your agent's eyes for video

A hosted MCP server, spoken natively by Claude, Cursor, LangGraph and any MCP-capable client. Add one URL — no install, no sidecar process.

get_video_transcript(url) exposed as a native tool
Streamable HTTP, stateless — nothing to babysit
Same key and quota as the REST API

mcp.json

{
  "mcpServers": {
    "video-transcript": {
      "url": "https://api.payload.cool/mcp/",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}

Metered, monthly, no surprises

Subscribe on RapidAPI — keys, billing and quotas are handled there, and you can cancel anytime. A request is one video, any length up to thirty minutes.

Free

25 requests a month

Full API access, every platform
Timestamped segments
Hard limit — no card required

Try it free

Pro · recommended

$9.90 / mo

300 requests a month

5¢ per extra request
Email support
Enough for a research-tool MVP

Ultra

$29 / mo

2,000 requests a month

3¢ per extra request
Priority queue
For apps in production

Mega

$79 / mo

8,000 requests a month

2¢ per extra request
Priority support
Beyond this — email us

Client errors (4xx) count toward quota; server faults (5xx) never do.

Questions engineers ask

How accurate is the speech recognition?

A production-grade model tuned for social audio — slang, names and profanity come through intact. When a platform ships human captions we prefer those (route: "captions"), so you always get the best available source.

What about latency?

Short-form video typically returns in 5–15 seconds; a 20-minute talk takes about a minute. The request is synchronous — one call, one JSON. The ceiling is 240 seconds.

Which languages?

English is the primary target and the best tested. The engine auto-detects major European and Asian languages; captions pass through in their original language.

Why is YouTube "coming soon"?

YouTube aggressively blocks datacenter traffic. Rather than ship something flaky, the API returns an explicit youtube_temporarily_unavailable today, and YouTube goes live once it meets the same reliability bar as everything else.

Do you store the videos?

No. Media is streamed, transcribed and discarded. Request logs — URL, duration, status — are kept for billing and abuse prevention; the content is not.

Is there an SLA?

Not yet — this is a young, honestly-priced service that would rather earn trust than print a badge. Server faults never count against your quota, and incident history is available on request.