Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use Haunt MCP to extract structured JSON, clean Markdown, article fields, metadata, or usage from permitted public web pages.
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 209% | 0% |
| case-09 | ✓→✓ | ＝ Same ✓ | 60% | 0% |
| case-10 | ✓→✓ | ＝ Same ✓ | -8% | 0% |
| case-11 | ✓→✓ | ＝ Same ✓ | -39% | 0% |
Use this skill when the user needs data from a permitted public web page inside Claude Code, Cursor, Codex, OpenCode, or another MCP-aware coding agent.
If the Haunt MCP server is installed but no API key is configured yet, call try_demo_extract first. It proves the server is wired correctly and returns docs, signup, pricing, MCP info, and free-tier links without using credits.
For live pages, require HAUNT_API_KEY in the MCP server environment. Use:
extract_url or extract for structured JSON from a permitted public URL.extract_markdown for readable page text for RAG, notes, docs ingestion, or .md files.extract_article for public article pages.extract_metadata for titles, descriptions, Open Graph, Twitter Card, and canonical URL metadata.get_usage after live calls to check plan and remaining credits.Haunt is for permitted public pages and supported rendered pages, including many that sit behind Cloudflare or a bot wall. It does not promise CAPTCHA solving, login-wall access, paywall access, or restricted-page access. If a page is blocked, login-required, CAPTCHA-gated, paywalled, restricted, or too thin to verify, treat Haunt's clear failure signal as the correct result. Do not fabricate extracted data.
"Use Haunt to extract the product name, price, availability, and review count from this permitted public URL: <url>. Return JSON and tell me if the page was blocked or unverifiable."
Other measured skills in the registry, with their headline benchmark lift.