What is MCP hosting? Deploying a site via MCP

MCP hosting means deploying and managing a website directly through the Model Context Protocol, so an AI agent like Claude, Cursor, or Windsurf can publish your site to a live URL without you uploading anything by hand. In practice: you connect a hosting provider's MCP server to your AI client once, then you tell the agent "deploy this site" in plain language and it ships the files, returns the public URL, and can read, edit, and redeploy later. VibeDeploy exposes exactly this at https://mcp.vibedeploy.be/mcp.

Below is what MCP is, why deploy-via-MCP is genuinely different from a normal deploy, and the exact setup.

What is MCP?

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in late 2024 and now supported across many AI clients, that lets an AI assistant call external tools through a consistent interface. Instead of an assistant only generating text, an MCP server gives it real actions it can take: query a database, read a file, or, in our case, deploy a website.

An MCP server publishes a set of named tools with descriptions. The AI reads those descriptions and decides which tool to call based on what you asked. When you say "put this online", a well-described deploy tool is what the agent reaches for.

What "MCP hosting" adds over a normal deploy

A normal deploy means you build your site, open a dashboard, and upload, or you push to Git and wait for CI. MCP hosting collapses that into the conversation you are already having with your AI tool:

• No context switch. You stay in Claude or your editor. The agent deploys; you do not alt-tab to a dashboard.
• No re-typing steps. The agent already has your files. It calls the deploy tool with them directly.
• Round trips. The agent can read the live site back, make an edit, and redeploy, all in one chat. "Fix the nav and ship it" becomes a single instruction.
• Plain language. You never learn a CLI or a YAML format. The tool descriptions do the matching.

This is why deploy-via-MCP fits AI-built sites so well: the same agent that wrote your site can put it online.

How to deploy a site via MCP with VibeDeploy

The setup is one-time. After that, deploying is a sentence.

Step 1: get a VibeDeploy account

Start a 14-day free trial, no credit card. This gives your agent something to authenticate against.

Step 2: connect the MCP server in your AI client

Add the VibeDeploy MCP server, https://mcp.vibedeploy.be/mcp, to your client. In an editor like Cursor or Windsurf, that is an entry in the MCP settings:

{
  "mcpServers": {
    "vibedeploy": { "url": "https://mcp.vibedeploy.be/mcp" }
  }
}

A browser window opens once so you can authorise it against your VibeDeploy account. This uses OAuth, so your agent gets a scoped token, not your password.

Step 3: tell the agent to deploy

Ask in plain language:

Deploy this site to VibeDeploy.

The agent calls the deploy tool, uploads your build, and replies with the live URL at yourname.vibedeploy.be. To iterate later: "read the homepage, fix the footer link, and redeploy." It happens in-chat.

Step 4: add a custom domain (optional)

Ask the agent to add your domain, or do it in the dashboard. Point the DNS records it gives you and SSL provisions automatically.

Is deploy-via-MCP safe?

Yes, with the usual care. VibeDeploy's MCP server uses OAuth 2.1 with a real consent screen, so you approve access explicitly and the agent receives a scoped, revocable token. Destructive actions are clearly marked and the most dangerous one, replacing every file on an existing site, requires an explicit confirmation. You can revoke the connection at any time from your account.

What MCP hosting does not do

It does not turn a static host into a backend platform. VibeDeploy serves static and single-page sites; if your project needs an always-on server, that layer lives elsewhere. MCP hosting also does not replace your editor or your AI tool. It is the publish step, wired into the tools you already use.

Where this fits

Deploy-via-MCP is one of three ways to ship to VibeDeploy, alongside dashboard upload and Git build-on-push. If you do not use an AI agent, the other two work exactly as well. See localhost to live URL for the manual path, the tool-by-tool guides for your specific client, and the pillar guide, How to deploy AI-built websites in 2026, for the full picture.

Ready to try it? Start free, connect the MCP server, and tell your agent to deploy.