npx -y --package @hauntapi/cli@latest haunt-cli init
Run this once to print the MCP config for Claude, Cursor, Windsurf, or another MCP-compatible client.
View the docs >Use the MCP command when your agent supports tools. Agents can extract structured JSON or clean Markdown from permitted public pages. Use the curl skill when an agent needs setup instructions first.
npx -y --package @hauntapi/cli@latest haunt-cli init
Two clean entry points. No wall of config unless somebody actually asks for it.
npx -y --package @hauntapi/cli@latest haunt-cli init
Run this once to print the MCP config for Claude, Cursor, Windsurf, or another MCP-compatible client.
View the docs >curl -sS https://hauntapi.com/agents/skill.md
Give this to an AI agent so it can read the Haunt setup path before it starts calling tools.
View the skill >try_demo_extract before getting a key or spending credits.HAUNT_API_KEY, then treat it like a password.extract or extract_url for JSON, or extract_markdown when the workflow needs clean page text or a .md file.device to mobile or desktop for the right rendering, or use js_scenario (paid) to wait, click, scroll, or fill before extracting.get_usage to see used, reserved, and remaining credits.diagnostics object (reason, retryable, suggested_action) instead of guessing. It covers captcha, login_required, access_denied, and more.For hosted MCP clients, use https://hauntapi.com/mcp/server and send Authorization: Bearer ${HAUNT_API_KEY}.
For local stdio MCP, use @hauntapi/mcp-server. The package exposes the haunt-mcp binary plus try_demo_extract, extract/extract_url, extract_markdown, extract_article, extract_metadata, and get_usage tools.