mcp web extraction

Give your AI agent a web extraction tool.

Start with the no-key try_demo_extract tool. Add a free Haunt key when you want the agent to extract real reachable public pages.

Try no-key demo Get free key

Haunt does not solve CAPTCHAs, log in for you, or bypass verification pages. It returns clear failures when a page is blocked.

hosted MCP config

{
  "mcpServers": {
    "haunt": {
      "url": "https://hauntapi.com/mcp/server",
      "headers": { "Authorization": "Bearer ${HAUNT_API_KEY}" }
    }
  }
}

tool 1

`try_demo_extract`

No key required. Returns fixed sample JSON so your MCP client can prove the tool connection first.

tool 2

`extract`

Requires HAUNT_API_KEY. Sends URL plus prompt and returns success, data, and trace.

tool 3

`get_usage`

Requires a key. Shows plan, monthly successful-request limit, reserved usage, and remaining quota.

local stdio MCP

For Claude Desktop, Cursor, and local MCP clients.

Use the public npm package. The demo tool works first; real extraction needs your key in the environment.

claude_desktop_config.json

{
  "mcpServers": {
    "haunt": {
      "command": "npx",
      "args": ["-y", "@hauntapi/mcp-server"],
      "env": { "HAUNT_API_KEY": "${HAUNT_API_KEY}" }
    }
  }
}

free key

Create a key, then paste it into MCP.

Free includes 1,000 credits per month. Failed extractions do not burn credits. No automatic overages.

prefer http?

REST is still the simplest app integration.

If you do not need MCP, call the REST endpoint directly with X-API-Key.

first REST extraction

curl https://hauntapi.com/v1/extract \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $HAUNT_API_KEY" \
  -d '{"url":"https://example.com","prompt":"Extract the page title"}'