Skip to main content
/v1/extract is an async AI operation that watches the whole video and returns a structured answer to your prompt. POST to enqueue the job, then poll until the result is ready. If you’re using the MCP server, polling is handled automatically and your agent receives the finished result directly. Credits: 10 credits per extraction.

​
Step 1 — Enqueue the job

This endpoint requires an x-api-key header and Content-Type: application/json.

​
Body parameters

​
string
required
The full public video URL, or a direct media file URL.
​
string
required
What to extract or analyze — for example: "List all product names", "Summarize in 3 sentences", or "What is the hook?".
​
string
Optional language hint for transcription — e.g. "en", "es", "ur".
​
object
Optional JSON Schema object that constrains the shape of resultJson. Use this to get typed, predictable output that slots directly into your application.

​
Example

​
Response

​
Step 2 — Poll for the result

​
Example

​
Response when complete

Poll until status is "completed" or "failed".

​
Example prompts

The prompt field is free-form. Here are prompts that work well:
  • "Extract the hook, main argument, and call to action"
  • "List every product mentioned with the timestamp it appears"
  • "What claims does the speaker make? Rate each one as factual, opinion, or unverified"
  • "Rewrite this as a Twitter thread"

​
Using JSON Schema for typed output

Pass a schema field to constrain the structure of resultJson. This is useful when you need the result to conform to a specific interface in your application.
If you’re using the Veedcrawl MCP server, polling is handled automatically. Your agent receives the completed extraction result without any extra logic.