I built Pageree, a remote MCP server that lets Cursor publish a landing page with a working form, then read how visitors use it. I sell it, so to be upfront: it’s a 30-day free trial with a card, then $5 to $40 a month, and the trial has to be running before the agent can save a draft.
Cursor can write a good page quickly. The work starts after that: hosting, a form that delivers leads somewhere, spam checks, a privacy policy and analytics. Pageree covers those through one connection.
Setup
The docs have an Add to Cursor button, or you can add it by hand to ~/.cursor/mcp.json:
{
"mcpServers": {
"pageree": { "url": "https://mcp.pageree.com/" }
}
}
Then open Cursor’s MCP settings, find pageree and authenticate. You sign in through the browser, and there’s no key to paste.
What you can ask Cursor to do
- “Build a landing page for my bike repair shop with a booking form.” You get a private preview link, and nothing goes live until you approve it.
- “Send the leads to my email.” They can also go to a CRM through a webhook, and any keys go in our console, not in the chat.
- “How is the page doing?” It reads visits and sources, plus where people click, stop and leave, section by section, then proposes one change. Every publish is a version you can roll back.
Two things I learned designing the tools
- A tool description isn’t a guardrail.
publishtells the agent to wait for approval, but an agent still called it right aftersave_draft. So the server now refuses a publish within 30 seconds of the last save. - Long instructions live in read-only tools. The markup and form rules don’t fit in tool descriptions, so the agent loads tools like
get_skill_page_structurewhen it needs them.
What would you want an agent to handle after a page is live?