Page to JSON — Structured Field Extraction for AI Agents

Give a URL and the fields you want. Get back strictly those fields as clean JSON.

Page to JSON is a small, fast, agent-native extraction primitive. You pass a page URL (or a list) plus a field list — either field names (price, sku, author) or short natural-language phrases (product image, site name) — and you get back exactly those fields as a flat JSON object. No markdown, no page dump, no prose. Token-minimal by design so an AI agent can call it and drop the result straight into a prompt or a database row.

Callable by AI agents via API/MCP, payable per call. Every field is a billable event only when it is actually filled — you never pay for a null.

Why it's different

Most "AI scraper" actors charge per page they fetch, whether or not your requested data actually came back, and many return a wall of markdown you still have to parse. Page to JSON is the opposite:

• Pay only for filled fields. The primary pay-per-event unit is field-filled (USD 0.005). A field with no confident signal is returned as null and is not charged. Ask for 6 fields, get 4 → you pay for 4.
• Strict output. The output object contains only the keys you asked for. Nothing extra to trim, nothing to hallucinate.
• Deterministic, not a guess. Values come from the structured data the page actually publishes about itself — schema.org JSON-LD, OpenGraph / product / Twitter-card meta tags, standard <meta>, <title>, canonical link, <h1>, and a few visible-text heuristics (email, phone, price). No LLM in the loop → no hallucinated values, no token cost, fast and cheap.
• Agent / MCP native. One URL + one field list in, strict JSON out. Trivial to wire into an agent tool, an MCP server, or an x402 pay-per-call flow.

Input

Field	Type	Description
url	string	A single public page URL. Use this or urls.
urls	array of string	Batch of public page URLs; one record per URL.
fields	array of string	The fields you want back (names or short phrases). Required. Output keys are strictly these.

{
  "url": "https://www.example.com/products/widget",
  "fields": ["title", "price", "currency", "brand", "sku", "product image"]
}

Output (one dataset item per URL)

{
  "url": "https://www.example.com/products/widget",
  "ok": true,
  "fields": {
    "title": "Aeropress Go Travel Coffee Press",
    "price": "39.95",
    "currency": "AUD",
    "brand": "Aeropress",
    "sku": "AP-GO-001",
    "product image": "https://.../aeropress-go.jpg"
  },
  "source_map": { "price": "jsonld:price", "brand": "jsonld:brand" },
  "filled": ["title", "price", "currency", "brand", "sku", "product image"],
  "missing": [],
  "filled_count": 6,
  "requested_count": 6
}

source_map gives provenance for every filled field so you can trust where each value came from.

Pricing

Pay-per-event:

• field-filled — USD 0.005 per requested field successfully extracted (primary).
• apify-actor-start — a fractional start fee per run.

You pay for results, not attempts. Unfilled fields cost nothing.

Good for

• AI agents that need a clean, typed field or two from a page without a full crawl.
• Product / price / availability lookups from a product URL.
• Article metadata (title, author, published date, image) for content pipelines.
• Any "URL → these exact fields → JSON" step inside a larger automation or MCP toolset.

Limits & honesty

• Extracts what a page publishes as structured data (JSON-LD/OpenGraph/meta/microdata) plus light text heuristics. It does not reason over arbitrary body prose like an LLM. If a page exposes no structured signal for a field, that field returns null (and is free). This is a deliberate trade for determinism, speed and near-zero cost per field.
• Public pages only — no proxy, no login-walled content, no headless browser.

Keywords: extract structured data, url to json, page to json, schema.org extractor, opengraph, json-ld, ai agent tool, mcp, structured extraction, pay per result, web data.

License: MIT.