Stop copying web
data by hand.
- Open pages, hunt for fields, paste into a sheet, and still miss things.
- Brittle scrapers shatter every time a layout shifts.
- Other tools fake success on blocked pages and hand you garbage.
- You inherit a pile of stored page content you never wanted to hold.
- A pricing page becomes a clean list of plans, prices, limits and features.
- Need page text for an agent or RAG pipeline? Ask for Markdown instead.
- One endpoint handles fetch, JavaScript render, and extraction.
- Blocked or login-only? Haunt says so, and never invents data.
- Page content is never stored, and results are kept only if you opt in.
Three moves,
no guesswork.
Try the demo, grab a free key to test a page of your own, then connect Haunt to your app or AI agent.
Run the live demo
See structured JSON and a confidence trace before you hand over an email. No signup, no key.
Claim a free key
Get 1,000 free credits and a ready-to-run curl command pointed at your own URL.
Build the integration
Switch between REST, MCP or the SDK and start consuming clean data in your app or agent.
Pick your lane.
Most teams start with one concrete extraction, then add the use cases that are actually theirs.
Catch price changes faster
Watch a lot of changing pages and get back the fields you can trust, every run.
Pricing monitorSummarise a company fast
Company summaries, metadata and competitor context pulled from public pages.
Company extractionGive your AI cleaner facts
Feed agents reliable website facts as structured JSON or Markdown instead of raw HTML.
AI agent pathIt saves time because it tells the truth.
Bad extraction tools waste your time by pretending blocked pages worked. Haunt is built to fail clearly when a page can't support the answer.
- Works on normal public pages: company sites, pricing pages, docs, articles and directories.
- Renders many JavaScript pages. Heavy app shells, login walls and verification pages can still fail, and Haunt says so clearly.
- Returns a clear error when a page is blocked, empty or login-only.
- Never invents data just to look successful.
Start free.
Pay when it's useful.
Simple monthly credits. Failed extractions don't burn credits. No surprise overages.
These plans are the self-serve API: you build, you run. Want competitor prices watched for you instead? Price Watch, done for you, from £49 a month.
- JSON or Markdown output
- Schema-checked, or an honest failure
- You never pay for failed calls
- Everything in Free
- Residential proxy retries blocked pages
- Scripted browser steps
- Everything in Starter
- Your own headers and cookies (beta)
- Higher rate limit
- Everything in Pro
- Highest rate limit and concurrency
- Built for heavy daily volume
Get a free key. Run one command.
No credit card. Free gives 1,000 credits each month. Your key appears right after, your first call runs itself while you watch, and then you point it at a page of your own.
Haunt never stores fetched page content. Extracted results are stored only if you opt into caching with max_age, and only for as long as you ask. Submitted URLs and prompts are kept encrypted for up to 60 days for abuse prevention and billing evidence, then deleted automatically.
Questions normal
humans ask.
What does Haunt do?
It extracts useful fields from public web pages as structured JSON, or returns clean Markdown when your agent or RAG pipeline needs page text.
Can I try it without signing up?
Yes. Try the demo above first. Create a key only when you want to test your own URL.
What if the page is blocked?
Haunt detects many CAPTCHA and human-verification blocks and returns a clear error like captcha_required. It does not solve CAPTCHAs or bypass login walls.
How do credits work?
Simple monthly plans based on credits. Failed extractions do not burn credits. No automatic overages. Simple pages usually use 1 credit, normal structured extraction 2, browser-rendered pages 4, and heavy pages 8. Very large pages add 1 credit per ~10,000 tokens of page content (up to 3 extra), so a big page costs a little more than a small one.
Can developers still use MCP or REST?
Yes. REST supports response_format: "markdown", and the MCP package exposes extract_markdown for agent workflows.
Haunt your first page
in the next five minutes.
Run the demo, get a free key, make one real extraction. No credit card, nothing to install.