Hey PH, Danny here, founder. I built Haunt because every agent project I touched had the same glue: fetch, parse, wrangle a headless browser. And every scraping API billed me for the requests that failed. Two things I'd love your take on: 1. It's MCP-native: a hosted streamable-HTTP endpoint plus a local npm package, so your agent just gets a "read this page, return JSON" tool it can call itself. 2. Failed extractions don't burn credits. A genuinely blocked page returns a clear error, charges zero, and never invents fields, so your agent branches on the failure instead of hallucinating. Free 1,000 credits a month, no card, plus a no-key demo so you can see the output shape first. Honest limits: it's not a CAPTCHA-solver or a bypass tool, and heavy app shells can still fail (it tells you when they do). Would genuinely value feedback from anyone shipping agents.

Haunt

Web extraction for AI agents, pay only for working calls

Web extraction for AI agents, pay only for working calls