AI-powered JSON extraction

Structured data from any page

Skip brittle CSS selectors that break every redesign. Send a URL plus a prompt and get JSON back. Optional schema for strict field validation.

Works with your agent stack

ClaudeCursorCodexVS CodeGeminin8n

Pricing tables, product specs, and directory listings change often. Prompt-based extraction adapts when layout shifts.

Pricing pages

Pull plan names, prices, and feature lists from competitor pricing without maintaining selectors.

Product catalogs

Extract SKU, title, and availability from listing pages into your database schema.

We load and render the page, then run AI extraction against the visible content. JavaScript-heavy sites return complete data.

Plain-English prompts

Describe the fields you need. No XPath or CSS required for the first version.

Optional JSON schema

Pass a schema object when you need strict types and required fields in the response.

Bring your model key

Use your own model credentials when you want control over extraction cost and model choice.

01

Define prompt

POST url and prompt to /v1/json. Add schema when you need strict structure.

View docs →
02

Schedule refreshes

Re-run on a cron. Cache keeps unchanged pages free on repeat calls.

Pricing →
03

Combine with crawl

Extract JSON from every page in a site crawl for full catalog coverage.

Site crawl →

JSON extraction FAQs

Questions about prompt-based structured extraction.

View docs
001Do I need a schema?

No. A plain-English prompt is enough. Add schema when you need guaranteed field names and types.

002What model is used?

You can bring your own model key. CaptureWeb handles render and passes page content to extraction.

003How is this different from /scrape?

/json uses AI to interpret page content. Selector-based scrape is better when you have stable DOM paths.

Start extracting structured data

Plain-English prompts. Rendered pages. JSON out.