MCP web extraction

Give your AI agent a web extraction tool.

MCP means your AI tool gets a Haunt button it can press for web data. Run the agent install command, add your free key, then ask it to extract a page.

Start with the no-key demo to prove the connection. Use the free key when you want real extraction.

Get a free key
agent install command 
npx -y --package @hauntapi/cli@latest haunt-cli init
step 1

Try the demo tool.

try_demo_extract proves your MCP client can talk to Haunt before you add a key.

step 2

Paste your key.

Create a free key, replace ${HAUNT_API_KEY}, and keep the key private.

step 3

Ask the agent.

Ask for the fields you need. Haunt returns structured JSON or a clear blocked-page failure.

free key

Create a key, then run the agent install.

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

local MCP

Using Claude Desktop, Cursor, or local stdio?

Use the public npm package from the docs. Setup, every config variant, and troubleshooting live there.

Open MCP setup
REST fallback

Building an app or script?

REST is the fallback for apps and scripts, not the main AI-agent setup. Use the terminal-specific first-call examples in the docs.

Open REST first call
what to ask your agent

Give Claude or Cursor real web access.

Once Haunt is wired in as an MCP tool, ask your agent in plain language. It calls Haunt and returns clean JSON, or a clear failure it can reason about.

example

“Get the pricing plans off acme.com”

Returns plan names, prices, limits and features as structured JSON.

example

“Summarise this company page”

Returns a clean summary, metadata and key links without raw HTML.

example

“Is this page blocked?”

Haunt says so clearly and doesn't invent data, so your agent can fall back.

Haunt works on normal public pages. Heavy app shells, login walls and CAPTCHAs can fail. Haunt reports the failure honestly instead of fabricating fields.

go deeper

Related MCP & agent guides.

guide

MCP server for web scraping

Connect Claude, GPT or any LLM to live web data with structured extraction.

 guide

Web scraping fallbacks for MCP servers

Handling 403, 429, thin content and structured extraction failures.

 setup

Agent setup in one command

The MCP npx command and the agent-readable setup skill.

 docs

MCP setup in the docs

Local stdio MCP, configuration variants and troubleshooting.

 use case

AI agent web extraction API

Demo, free key, MCP, REST and clean failure handling.

 capabilities

What Haunt can extract

Honest capability guide and platform limits.