From a ChatGPT or Claude chat to a live European website

There is a moment in every chat-built website project where the interesting part ends and the awkward part begins. The AI has produced the pages, the copy reads well, the preview looks right. Then the conversation stalls on a question the chat cannot answer by itself: how does this become a real site, on a real domain, that stays up after the tab closes?

Most guides answer with a detour through developer tooling: install a CLI, create a repository, configure a pipeline. For someone who built the site by describing it in sentences, that detour is where projects die. The better answer is to let the same conversation finish the job. There are two ways to do that on VibeDeploy, and which one you use depends on what kind of AI client you are talking to.

Handoff one: the deploy guide, for plain chat tools

ChatGPT, Claude in the browser, Gemini and similar assistants can write code and call web APIs from within the conversation, but they have no built-in knowledge of any specific host. The deploy guide fixes that: it is a single instruction file that teaches the AI exactly how to publish to VibeDeploy, in one paste.

The flow looks like this:

1. You paste the VibeDeploy deploy guide into the conversation, alongside the site the AI already built.
2. The AI reads the instructions, packages the files, and calls the deploy endpoint itself.
3. It replies with a live URL on EU infrastructure, HTTPS already active.

No account juggling mid-conversation, no copying files out of a code block into a folder structure you have to guess at. The tool-specific walkthroughs for ChatGPT, Claude artifacts and Gemini show the exact prompts, but the principle is identical everywhere: the guide is the bridge, the AI does the crossing.

One detail worth knowing: the guide deliberately tells the AI to omit optional parameters. Assistants like to be helpful by adding configuration nobody asked for, and the guide is written to prevent that.

Handoff two: MCP, for agent clients

If you work in Claude Desktop, Cursor, Windsurf or another client that supports the Model Context Protocol, the handoff gets tighter. Instead of pasting instructions per project, you connect the VibeDeploy MCP server once, at https://mcp.vibedeploy.be/mcp, and authorise it against your account through OAuth. From then on the agent has deployment as a native capability.

The difference in day-to-day use is iteration. With the deploy guide, publishing is an event at the end of a conversation. Over MCP, it is a verb inside it: the agent can list your sites, read what is live, apply an edit, and redeploy, all as steps in the same exchange. "Tighten the hero copy and ship it" is a complete instruction. The Cursor and Windsurf guides cover the two-line client configuration.

What happens on the European side

Either handoff lands in the same place. The files are deployed to EU data centres operated by a Belgian company, Serso BV, and served from an isolated environment per site. TLS certificates provision automatically. The platform subdomain is live immediately; a custom domain is one CNAME record away, and the AI can set that up too if you ask it.

For a site aimed at European visitors or customers, this matters beyond latency. The hosting layer, the deployment metadata and the account data all stay under EU jurisdiction, which means the compliance questions that follow a public launch, where is it hosted, who operates it, is there a DPA, have short answers. The background on why that is worth deciding deliberately is in data sovereignty for AI-built websites.

Iterating after launch

A chat-built site is rarely finished at first deploy, and the workflow holds up past launch:

• In-chat edits. Over MCP, the agent reads the deployed source, changes it, and redeploys. The site in production is the working copy.
• Staging. When the site starts to matter, changes can go to a staging copy first and be promoted when they look right.
• Snapshots. Every deploy can be rolled back, so an overconfident AI edit is a one-click recovery rather than an incident.

If your project lives on your own machine rather than in a chat, the same platform accepts a dragged-in build folder; that path is documented at deploy localhost to production.

Cost and the honest limits

Plans are flat: 15, 39 or 129 euro per month depending on scale, with a 14-day trial that needs no credit card. Details on the pricing page. The limit to be clear about is that this is static and single-page-app hosting. A chat-built frontend, marketing site, portfolio or dashboard shell fits perfectly; a project needing its own always-on server process does not.

The gap between "the AI built it" and "it is live in Europe" used to be the hard part of these projects. With the deploy guide for chat tools and MCP for agents, it is now the shortest step in the conversation.