If you stop paying
Functions are stored, not lost: they answer 402 until the account is paying again. The database works for seven days, is then stopped, and is never deleted; you are pointed at transfer_database, which has no deadline.
Tell your AI "put this site online", and you get back a link anyone can open.
On your phone? Connecting is done once, on a computer. Send yourself the guide for later.
Add it toClaude CodeCodexCursorVS CodeClaude Desktop, claude.aiChatGPT
"Publish this site." Nothing to configure — your assistant is already holding the tools.
Files, server code and a Postgres database go up together. The reply is the address.
Ask for an edit and it goes live again. Every earlier version is kept.
The simplest way: paste this line into your AI chat.
Or click through it yourself, with a picture of every screen:
claude plugin marketplace add https://juudd.com/marketplace.json
claude plugin install juudd@juuddThe plugin adds the connector above and the rules its tools enforce, in one step.
[mcp_servers.juudd]
url = "https://mcp.juudd.com/mcp"then codex mcp login juudd
{
"mcpServers": { "juudd": { "url": "https://mcp.juudd.com/mcp" } }
}{
"servers": { "juudd": { "type": "http", "url": "https://mcp.juudd.com/mcp" } }
}With GitHub Copilot
Customize → Connectors → Add, then paste the URL and keep what Claude detects.
Works on every Claude plan, Free included: Free allows one custom connector. A config file will not work for these.
Turn on Developer mode in Settings → Security and login, then Plugins → + → New Plugin with the connector URL.
Pro, Plus, Business, Enterprise or Education, on the web. Measured end to end on 7 September 2026.
Add a remote MCP server over HTTP at the connector URL.
The server advertises its own authorization metadata, so there is no API key, no token and no header to set by hand.
A framework project has two doors: build it yourself and post the folder, or send the source and the platform builds it in a sandbox.
A build runs against a toolchain this platform pins, and a package.json naming anything outside it is refused with the whole list in the reply. What a build cannot produce becomes a function.
site_errors says which routes failed, when and how many times, not why. What a request returns is all you have to debug from, so write your functions to answer with what went wrong.
Every version you deploy stays, so a site can be replaced but never wiped clean. Taking one off the internet is a different thing and it is allowed: unpublish_site stops serving it, and keeps every version.
On the paid tier your own domain works, but example.com with no www works only if your DNS provider flattens CNAMEs or offers ALIAS records. If yours does not, use www and redirect.
Ownership is proven through the domain's DNS, and another CDN in front of it hides exactly what has to be read. It has to come off that CDN first.
Priced and not yet purchasable: the billing that would collect it is not built.
The price works because this platform never pays for AI. You already pay Claude or ChatGPT; we give what it writes an address.
For static sites.
Freepermanently, not a trial
What you get:
For sites that need a database or a domain.
$12/ month
Everything in Free, plus:
A bot flood is traffic you did not want and cannot prevent, so it is not traffic you pay for. That is part of the price, not a support policy.
Why we do not count requestsFunctions are stored, not lost: they answer 402 until the account is paying again. The database works for seven days, is then stopped, and is never deleted; you are pointed at transfer_database, which has no deadline.
Add the connector, then describe what you want. Everything it will refuse is written above.
Connect your assistant