Visible content
public html excerpt
<div class="price-card"> <h3>Starter</h3> <div>£19/mo</div> <div>5,000 successful requests / month</div> </div>
before and after
Proof with a real page shape.
Example prompt: Extract the pricing plans as JSON with plan name, monthly price, request allowance, and included features.
Structured result
json output
{
"plans": [
{
"name": "Starter",
"price": "£19/mo",
"allowance": "5,000 successful requests / month",
"features": ["Public extraction"],
"notes": null
}
]
}boundaries
What works, what sometimes works, and clean failure.
Works well
Public marketing pages, docs pages, pricing pages, simple product pages, visible tables, and public JSON or XML endpoints.
Sometimes works
JavaScript-rendered public pages when the rendered content is visible to a browser session.
Clean failure
Login-required pages, CAPTCHA or human verification, private pages, thin public shells, and blocked targets. Haunt reports the failure class, it does not invent data.
api first
Run one real call.
REST call
first call curl
curl -X POST https://hauntapi.com/v1/extract \
-H "Content-Type: application/json" \
-H "X-API-Key: PASTE" \
-d '{"url":"https://example.com","prompt":"Extract the page title and pricing plans as JSON"}'Need a deeper walkthrough: open first-call docs.
Agent route
Using an MCP client instead of direct REST? Start from MCP server for web scraping and keep the same prompt shape.
More extraction recipes: recipes, capability guide.