Web Structured Data Extractor (Claude)

Pass a URL + JSON schema (or natural-language goal). Claude reads the page and returns a strict JSON object matching your schema. Product / news / hotel / real-estate / job-board extraction. BYO Anthropic API key. $0.01 per page.

---

Why this exists

You want to scrape structured fields out of arbitrary web pages — price, SKU, rating, hours, contact info, reviews. Building a per-site CSS-selector scraper is brittle (sites change every week). General-purpose LLM extraction is robust but requires prompting + parsing.

This actor wraps the whole pipeline:

1. URL → clean Markdown via trafilatura
2. Markdown + your schema/goal → Claude
3. Claude returns strict JSON → we parse and validate

Same idea as DiffBot's Article API ($299/mo) or Browse AI's extraction ($99/mo) — but with your own Claude API key and no monthly subscription.

---

What you get

{
  "url": "https://...",
  "model": "claude-opus-4-7",
  "goal": "Extract product info",
  "schema_used": true,
  "extracted_data": {
    "name": "Nintendo Switch 2",
    "price_usd": 499.99,
    "in_stock": true,
    "rating": 4.8,
    "reviews_count": 1247
  },
  "raw_output": "...",
  "input_chars": 5230,
  "usage": {"input_tokens": 1450, "output_tokens": 80}
}

---

Two ways to specify what to extract

Option 1: JSON Schema (recommended)

{
  "url": "https://www.amazon.com/dp/B07VPHN6CR",
  "schema": {
    "type": "object",
    "properties": {
      "name":         {"type": "string"},
      "price_usd":    {"type": "number"},
      "in_stock":     {"type": "boolean"},
      "rating":       {"type": "number"},
      "reviews_count":{"type": "integer"}
    }
  },
  "anthropicApiKey": "sk-ant-..."
}

Option 2: Natural language goal

{
  "url": "https://...",
  "goal": "Extract product name, USD price, in-stock status, average rating",
  "anthropicApiKey": "sk-ant-..."
}

You can combine both — schema sets the shape, goal sets emphasis.

---

Use cases

1. E-commerce competitor monitoring — Track price + availability across competitors
2. Real estate listing aggregation — Extract beds/baths/price/sqft from Zillow, Redfin, Realtor
3. Job board scraping — Title, company, salary, location, remote-flag from LinkedIn, Indeed
4. News article fact extraction — Get the same 5 fields from any news source
5. Hotel / travel research — Name, rating, price/night, amenities from any booking site

---

Pricing

Pay-Per-Event: $0.01 per page (Apify-side).

Anthropic tokens charged separately. Typical:

Page complexity	Input tokens	Anthropic	Total
Simple product page	~1500	~$0.008	$0.018
Long article	~4000	~$0.020	$0.030
Big e-commerce listing	~8000	~$0.040	$0.050

Use Haiku for batch / cheap extraction (~10x cheaper).

---

Setting your Anthropic API key

See Article Summarizer README for the full BYO API key guide. Short version:

1. Get key at console.anthropic.com
2. Paste in anthropicApiKey input (Apify saves it encrypted)
3. Or save as Apify Account-level Secret and reference as @MY_KEY

---

Tips for reliable extraction

• Better prompts → better results. A goal string like "extract product name, USD price, in-stock boolean, rating 1-5" outperforms "extract product info".
• Constrain types in the schema. "type": "number" is stricter than "type": ["string","number","null"].
• Test with Haiku first. Haiku 4.5 is fast and 10x cheaper for prototyping. Switch to Opus 4.7 when you need accuracy.

---

Related actors (same author)

• Article Summarizer — TL;DR instead of structured
• Web Page → Markdown Converter — Just the body, no LLM
• HTML Metadata Extractor — Cheaper for OG / Twitter / JSON-LD
• JSON Schema Generator — Bootstrap a schema from samples

---

Feedback

A short review helps engineers find it: Leave a review on Apify Store