How is this different from Firecrawl's JSON format?

Same idea: a JSON Schema and an optional prompt per page. Both bill the extraction on top of the page read (SerpKite +4 credits per page, Firecrawl +4). SerpKite also sends boolean and enum schemas to a decision model that always answers within the schema, bills only the page when an extraction fails, and sells credits that never expire. See SerpKite vs Firecrawl.

Can it extract from a whole site?

Map or crawl the site first, then send the pages you need to /v1/extract with your schema, 20 at a time.

What if a field isn't on the page?

Make it nullable in your schema, and say so in the prompt ("null when missing"). The model is told to use only what the page says.

Is the page content kept?

Pages are cached for up to an hour like any /v1/extract result; the content sent to the model is not logged or used for training.

POST /v1/extract · output_schema

Structured data extraction

Stop writing a parser per site. Describe the fields you want as a JSON Schema, send the URLs, and get back JSON that matches it, written by Claude from each page's content.

Get a free API key Structured extraction guide

How it works

Pages and PDFs to JSON that matches your schema.

output_schema on /v1/extract fills your JSON Schema from each page: up to 20 URLs or PDFs per call, fetched in parallel, logged out, with JavaScript pages rendered when needed. An optional prompt says what to pick and how to normalise it. Claude Haiku 5.5 writes the JSON, and values come only from the page.

A schema made only of booleans and enums is a classification, and goes to OpenAI's Decisions API instead, which always answers with a value the schema allows. summary: true adds a short summary of each page (+4 credits).

Each page costs 1 credit plus 4 for the JSON. A page that fails is listed in failed and costs nothing; an extraction that fails is reported in output_error and only the page is billed. Short pages that turn out to be bot checks, login walls or soft 404s are not billed at all.

Extract a pricing table as JSON

cURL

TypeScript

curl https://api.serpkite.com/v1/extract \
  -H "Authorization: Bearer $SERPKITE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"urls":["https://www.example.com/pricing"],"output_schema":{"type":"object","properties":{"plans":{"type":"array","items":{"type":"object","properties":{"name":{"type":"string"},"price_per_month":{"type":"number"}}}}}},"prompt":"Monthly prices in USD"}'

results[].output · illustrative

{
  "url": "https://www.example.com/pricing",
  "output": {
    "plans": [
      {
        "name": "Starter",
        "price_per_month": 19
      },
      {
        "name": "Team",
        "price_per_month": 79
      }
    ]
  }
}

What each page can return

What each page can return
Option	Returns	Credits per page
(default)	`markdown`, `text` or `html` with metadata	1
`query` + `highlights`	The passages that match the query (BM25)	1
`summary: true`	A 3–6 sentence summary	1 + 4
`output_schema` (+ `prompt`)	JSON that matches your schema	1 + 4

Why it matters

What you get

Your schema, your fields

Up to 16 KB of JSON Schema: nested objects, arrays, enums and nullable fields.

Schema-valid classification

Boolean and enum schemas are answered by a decision model that can't step outside the schema.

HTML and PDF

Product pages, docs, filings and papers, 20 per call, fetched in parallel.

Pay for what worked

Failed pages are free, and a failed extraction costs only the page read.

FAQ