--- title: "Supadata Integration" description: "Supadata is an API for video transcripts, social media metadata, AI video content extraction and web scraping across YouTube, TikTok, Instagram, X, Facebook and any public URL. Agents fetch and translate transcripts, list channel and playlist videos, scrape and crawl web pages and extract structured data from video." url: https://flowrunner.ai/integrations/supadata date_modified: 2026-09-04T14:59:52-07:00 --- # Supadata [Developer Tools](https://flowrunner.ai/integrations/category/developer-infrastructure) Supadata is an API for video transcripts, social media metadata, AI video content extraction and web scraping across YouTube, TikTok, Instagram, X, Facebook and any public URL. Agents fetch and translate transcripts, list channel and playlist videos, scrape and crawl web pages and extract structured data from video. [Verified](https://flowrunner.ai/integrations/verified "What does verified mean?") · 21 actions · API key · available [Supadata website](https://supadata.ai/) · [Platform Documentation](https://docs.supadata.ai/) · Capability data verified 2026-08-27 1. A weekly market review starts on a schedule 2. List Channel Videos returns the ids each tracked channel published in the window 3. Get Account Information reads the remaining credit allowance against that video count 4. The agent compares the real video count with the batch limit it is about to send 5. The agent posts the scope, the transcript mode, and the projected cost to the analyst 6. The analyst confirms the scope before Create Transcript Batch spends against it ## What This Integration Enables Supadata exists because the most useful things your market says are usually said out loud. Competitors announce positioning in launch videos, customers describe their real workflow in community streams, and analysts publish their view as a podcast. None of those platforms offers an API that hands you the words. Supadata's whole shape is that one job: give it a URL from YouTube, TikTok, Instagram, X, Facebook, or the open web, and get back text you can query, summarize, and route. FlowRunner agents use it in two directions. They pull transcripts and metadata for known sources, one video at a time or as batch jobs across a whole channel or playlist, and they discover sources through Search YouTube, Map Website, and Create Crawl Job. Extract Video Data goes further and analyzes what is actually seen and heard in a video, returning structured fields rather than prose. The agent handles the fanning out, the polling, and the shape normalizing. What FlowRunner adds on top is that the expensive and irreversible parts of that work do not run unattended, because Supadata is metered and because two of its defaults quietly change the answer without changing the status code. That is where [human-in-the-loop](https://flowrunner.ai/concepts/human-in-the-loop) earns its place in a research pipeline. ### Without FlowRunner **The research is unread**: Most of what a market says in public is spoken on video, and nobody has time to watch it **Transcription is a side project**: Pulling text out of a channel means a separate tool, a separate login, and manual copying **Findings arrive after the moment**: By the time a launch video has been watched and summarized, the response window has closed ### With FlowRunner **Spoken sources become searchable text**: Transcripts across YouTube, TikTok, Instagram, X, Facebook, and the open web land in the same shape **Collection runs as a background job**: Batches and crawls run against channels, playlists, and whole sites without anyone supervising them **Scope is a decision, not a default**: The analyst sees how much the run will cover and what it will cost before it starts ## Use Case Scenarios ### A competitive brief that reads what was said, not what was written Each week the agent walks a list of tracked competitor channels. List Channel Videos returns what was published in the window, Create Transcript Batch collects the text, and Get Batch Result polls until the job finishes. The transcripts go to a summarizing model such as [Anthropic AI](https://flowrunner.ai/integrations/anthropic-ai), and the resulting brief is written to [Notion](https://flowrunner.ai/integrations/notion) with links back to each source video. A short digest posts to [Slack](https://flowrunner.ai/integrations/slack) with the claims that changed since last week called out. The team reads positioning shifts in the week they happen rather than the quarter they happen. ### One long recording becomes a month of assets A webinar or conference talk finishes. The agent calls Get YouTube Transcript for the full text and Get YouTube Video for the title, description, and publication date. Translate YouTube Transcript produces the versions the regional teams need. The transcript is turned into a blog draft published to [WordPress](https://flowrunner.ai/integrations/wordpress), a set of pull quotes appended to [Google Sheets](https://flowrunner.ai/integrations/google-sheets) for the social calendar, and a summary attached to the campaign record in [HubSpot](https://flowrunner.ai/integrations/hubspot). One recording, done once, reaches every surface without anyone retyping it. ### Sizing a site before paying to read it An analyst wants everything a vendor publishes about a product line. The agent runs Map Website first, which returns every link on the site without fetching any page content, and reports back that the site has far more pages than the crawl budget covers. It proposes a filtered subset and asks for confirmation. Once the scope is agreed, Create Crawl Job fetches only those pages as Markdown, Get Crawl Result walks the output, and Scrape Web Page picks up anything added later. The cheap read decided the expensive one, and a person decided the boundary. ## Human-in-Loop Highlight Supadata's batch operations accept a whole channel or playlist instead of a list of ids, and when they do, the batch limit decides how many videos are actually processed. Its default is lower than most callers expect, and a run that stops at the default returns success. That is the dangerous shape: an analyst asks what a competitor has been saying all quarter, the agent transcribes the first handful of videos, and the brief that comes back reads exactly as confident as one built on all of them. The transcript mode has the same property in the other direction, where captions-only runs return unavailable on the older uploads and live streams that often carry the most candid material. So the agent stops before it spends. It posts: "Channel has published 214 videos in this window. The batch is currently limited to 10. Mode is set to native captions only, which will skip an unknown number of live streams. Raise the limit to 214 at the higher credit cost, keep the sample, or switch modes?" The analyst answers, and the number they see in the finished brief means what they think it means. FlowRunner is not slowing the research down. It is making sure the scope of the answer is a decision somebody made. Agent processes routinely Detects exception requiring judgment Clear match Continues automatically Ambiguous Routes to human via preferred channel Human decides Agent resumes with decision ## Agent Capabilities 21 actions ### Transcripts 5 - **Get Transcript** Returns the transcript of a video or audio file from YouTube, TikTok, Instagram, X, Facebook, or a public file URL. Its mode parameter decides whether the text comes from the platform's own captions, from AI transcription, or from captions with an AI fallback. - **Get Transcript Result** Returns the result of an asynchronous transcript job by id. Long media returns a job id rather than content, and this is the polling route for it. - **Get YouTube Transcript** Returns a YouTube video's transcript specifically, addressed by URL or by video id. - **Translate YouTube Transcript** Translates a YouTube transcript into another language. The target language is required here rather than being a preference. - **Create Transcript Batch** Queues transcription across many YouTube videos at once, addressed as a list of ids, a playlist, or a whole channel. Returns a job id only. The operation this page's human gate exists for. ### YouTube 8 - **Get YouTube Video** Returns a video's title, description, channel, duration, publication date, engagement counts, tags, and thumbnails. - **Create Video Batch** Queues metadata collection across many YouTube videos at once, from a list of ids, a playlist, or a channel. - **Get Batch Result** Returns the status and, once complete, the output of a batch job. Both batch types share this single polling route. - **Get YouTube Channel** Returns a channel's name, handle, description, subscriber and video counts, and imagery, from an id, a handle, or a URL. - **List Channel Videos** Returns the video ids a channel has published, optionally narrowed to regular videos, Shorts, or live streams. The scoping read that runs before a batch. - **Get YouTube Playlist** Returns a playlist's title, description, owning channel, and video count. - **List Playlist Videos** Returns the video ids in a playlist, ready to feed into a batch when full details are needed. - **Search YouTube** Searches videos, channels, and playlists using the filters the site itself offers, including upload recency, duration, result type, sort order, and features such as HD or subtitles. ### Web 4 - **Scrape Web Page** Fetches one page and returns its content as Markdown alongside the title, description, and outbound links. Markdown is what makes the result directly usable as model input. - **Map Website** Returns every link discovered across a site without fetching page content. The cheap way to size a site and choose what is worth crawling. - **Create Crawl Job** Queues a crawl that fetches every page on a site as Markdown up to a page limit, returning a job id. - **Get Crawl Result** Returns the status and pages of a crawl. Large crawls page their own output, so results are walked rather than expected all at once. ### Media 3 - **Get Media Metadata** Returns a post or video's metadata from YouTube, TikTok, Instagram, X, or Facebook in one shape that is the same across all of them, so a flow can handle mixed sources without branching. - **Extract Video Data** Uses AI to analyze what is seen and heard in a video and return it as structured data. This is content analysis rather than metadata, and passing a schema is what keeps the output shape stable between calls. - **Get Extract Result** Returns the status and structured output of an extraction. When a job ran from a prompt alone, the response also carries the schema the model generated, which can be reused to pin later calls to a fixed shape. ### Account 1 - **Get Account Information** Returns the organization, plan, and credit usage on the configured key. Called before any flow that fans out across a channel or a site. ## Frequently Asked Questions ### What can FlowRunner do with Supadata? FlowRunner agents can run Get Transcript, Get Transcript Result, and Get YouTube Transcript in Supadata, plus 18 more actions. ### Does connecting Supadata to FlowRunner require OAuth? No. Supadata connects to FlowRunner with an API key, no OAuth flow required. ### Can Supadata trigger a FlowRunner workflow automatically? Supadata doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger. **Work at Supadata?** This integration exposes Supadata to AI agents on every FlowRunner plan, including through MCP, at no cost to you. [See what FlowRunner offers integration partners](https://flowrunner.ai/integrations/partners), including how to keep this page current. --- Markdown version of https://flowrunner.ai/integrations/supadata. Site index: https://flowrunner.ai/llms.txt