SPLICE LOG

Tested tool · Transcript API

Supadata

SUPADATA, turns the spoken audio of a public video into clean text with one API call. If the video already has captions, they come straight back with timestamps. If not, it makes a transcript and hands you a job to check on.

Read the full spec

Sample entry. The test below was run by LARGE FORMAT with its open-source video2notes tool, and is adapted here for review.

Ask your AI about Supadata

Prompt

I'm reading the SpliceOps Splice Log entry for Supadata at https://spliceops.lrgfrmt.com/log/supadata/. Read https://spliceops.lrgfrmt.com/log/supadata.md and, in plain English, explain what Supadata is and how it could help my team. If you can't open the page, say so, then explain it from what you know.

brew install ffmpeg tesseract        # macOS
git clone https://github.com/lrgfrmt/video2notes.git
cd video2notes && python -m venv .venv && source .venv/bin/activate
pip install -e .
export SUPADATA_API_KEY=your-key
video2notes "https://www.youtube.com/watch?v=_20US068pzk" -o notes.md

MIT licence. Needs Python 3.10+, ffmpeg and tesseract.

pip install supadata

from supadata import Supadata
supadata = Supadata(api_key="YOUR_API_KEY")
t = supadata.transcript(url="https://youtu.be/_20US068pzk", text=True, mode="auto")

Job status checks need supadata 1.7.0 or newer.

# Claude or ChatGPT: add a custom connector with this URL
https://api.supadata.ai/mcp

# Cursor, Windsurf, VS Code (local)
env SUPADATA_API_KEY=your-key npx -y @supadata/mcp

Hosted server signs in with OAuth; the local one reads SUPADATA_API_KEY.

Stage: Added
Type: Transcript API
Logged: 01.11.2026
Controls: Spend cap / kill switch / full record

Next in the log
jev LG-118

Spec sheet · SupadataLG-119

Part / LG-119Stage / AddedChecked / 2026-11-01Rev / 1.0

01Overview

Why it's in the log

A lot of what a company knows lives in video: training clips, recorded demos, webinars and walkthroughs. It is hard to search and easy to lose. Supadata is the listening half of turning those videos into notes a team can find later.

What we found

Neither half is enough on its own. In the sample video the captions heard a name as “JJ Lair”. The title card on screen spells it right: JJ Allaire. Code, prices, links and product names are usually shown, not said. Put the spoken and on-screen tracks on one clock and the note becomes usable. An 86-second video took about 25 seconds end to end.

What it means for your team

Recorded training and walkthrough videos can become searchable notes in the places your team already keeps what it knows. A person still decides what gets filed and shared. [How this fits a specific client workflow is for the working session.]

Spoken (captions)

“JJ Lair”

On screen (title card)

JJ Allaire

02Specifications

Maker: Supadata (supadata.ai)
Category: Transcript and web-data API
Base URL: https://api.supadata.ai/v1
Authentication: Header x-api-key
Video sources: YouTube, TikTok, Instagram Reels, X, Facebook, public file URLs
File formats: MP4, WEBM, MP3, FLAC, MPEG, MPGA, M4A, OGG, WAV
File limits: 750 MB and 12 hours per file
Transcript modes: native (existing captions), generate (AI), auto (native, then generate). Default auto
Long videos: Over 20 minutes returns a job ID (HTTP 202) to poll
Job results: Kept for 1 hour after completion
SDKs: Python: supadata. Node: @supadata/js
MCP server: Hosted at https://api.supadata.ai/mcp (OAuth), or local @supadata/mcp
No-code: n8n, Zapier, Make, Activepieces
Not supported: Private, login-only, age-restricted or heavily geoblocked videos, and live streams

03Endpoints

Method	Path	What it does
GET	/v1/transcript	Transcript for a video or file URL
GET	/v1/transcript/{jobId}	Result of a long or generated transcript
GET	/v1/metadata	Post and video metadata
POST	/v1/extract	Structured data from a video, using AI
GET	/v1/web/scrape	A web page as markdown
GET	/v1/web/map	All links on a site
POST	/v1/web/crawl	Crawl a site (Pro plan and up)
GET	/v1/youtube/…	Search, channel, playlist, batch and translation

Paths for metadata, extract, web and YouTube follow the docs' structure; check the API reference before building against them.

04Install and connect

Splice fork · members

# Sign in to get the Splice fork address.

Original tool

brew install ffmpeg tesseract        # macOS
git clone https://github.com/lrgfrmt/video2notes.git
cd video2notes && python -m venv .venv && source .venv/bin/activate
pip install -e .
export SUPADATA_API_KEY=your-key
video2notes "https://www.youtube.com/watch?v=_20US068pzk" -o notes.md

Python SDK

pip install supadata

from supadata import Supadata
supadata = Supadata(api_key="YOUR_API_KEY")
t = supadata.transcript(url="https://youtu.be/_20US068pzk", text=True, mode="auto")

MCP

# Claude or ChatGPT: add a custom connector with this URL
https://api.supadata.ai/mcp

# Cursor, Windsurf, VS Code (local)
env SUPADATA_API_KEY=your-key npx -y @supadata/mcp

05Pricing

Plan	Price a month	Credits a month	Requests a second
Free	$0	100	1
Basic	$5 a month, billed yearly	300	10
Pro	$17	3,000	10
Mega	$47	30,000	50
Giga	$297	300,000	100
Supa	$897	1,000,000	100
Enterprise	Custom	Any	Custom

06What a credit buys

Action	Cost
Transcript (existing captions)	1 credit
Generated transcript	2 credits per minute
Web scrape	1 credit per page
Translation	30 credits per minute
Extract	5 credits per minute, 5 minimum
Unused credits	Do not roll over

07Error codes

Code	Meaning
206	Transcript unavailable for this video
400	Check the request
401	API key missing or invalid
402	Endpoint not on your plan
403	Private or restricted content
404	Not found, or job result expired
429	Rate or usage limit reached
5xx	Error on Supadata's side

08Limits and caveats

Live streams are not supported.
Requests that detect no speech still bill by media duration.
Generated transcripts can take up to about 100 seconds before a job ID is returned.
The free plan allows 1 request per second and no advanced endpoints.
On-screen text reading in video2notes can misread stylised fonts, motion blur and busy backgrounds.

09How we tested it

Listen

Supadata's Python SDK fetches the transcript. Generated transcripts come back as a job, polled until done.

Read

ffmpeg keeps a frame at every scene change, Tesseract reads it, and only lines at least 75% certain are kept: 573 lines became 220.

Merge

Spoken and on-screen text sit on one clock in one markdown note. Repeats appear once, when they first show.

10Questions people ask

What is Supadata?

Supadata is an API that returns transcripts and metadata for videos on YouTube, TikTok, Instagram, X and Facebook, and for public video or audio files. It can also scrape, map and crawl websites as markdown.

Can Supadata transcribe videos that have no captions?

Yes. In generate or auto mode it creates a transcript with AI, billed at 2 credits per minute of video. Native mode only returns captions that already exist.

How much does Supadata cost?

There is a free plan with 100 credits a month. Paid plans start at $5 a month billed yearly (300 credits) and $17 a month for 3,000 credits. One existing transcript costs 1 credit.

Does Supadata work with Claude and ChatGPT?

Yes. Its hosted MCP server at https://api.supadata.ai/mcp can be added as a custom connector in Claude and ChatGPT, and signs in with OAuth.

How does Splice Labs use Supadata?

[To confirm with the Splice Labs team.] In this sample it is the transcript step of a video-to-notes pipeline, run behind the same controls as every SpliceOps tool: a spend cap, a kill switch and a full record.

11Sources

Last checked 2026-11-01. Facts can change; check the sources before you build.

spliceops.lrgfrmt.com/log/supadataAdded / 01.11.2026