Give Claude or Cursor a cloud sandbox, a database and a publish button over MCP

Coding agents in Claude, Cursor and Codex are good at writing code. They're weaker at everything around it: a machine to run the app on, a live preview, a database, file storage, secrets and a public URL at the end. Usually you wire that up yourself.

This post shows how to hand all of that to the agent through one remote MCP server. I build Mythex, an AI app builder from Dubai, and this is how its MCP server works. The pattern applies to any MCP server that exposes a sandbox.

What the agent gets

Once connected, the client can:

• create or open a project and wake a cloud sandbox
• read, write, edit and search files, and run bash
• start a preview and read its logs (set_run_config, start_preview, get_preview)
• set secrets you provide (keys are listed back, values never are)
• add a Postgres database and file storage, make checkpoints, and publish
• run a Playwright browser script and get screenshots back in chat
• export or import code, sync with GitHub and connect a domain

Every tool result includes a workspace_url, so you can watch the live preview in the browser while the agent works in your editor.

Connect from Claude

1. Copy the MCP URL: https://mcp.mythex.ai/mcp
2. Claude → Settings → Connectors → Add custom connector → paste the URL → Connect
3. A browser tab opens the consent page. Sign in and click Allow access.

Connect from Cursor

Add the server with only the URL (.cursor/mcp.json):

{
  "mcpServers": {
    "mythex": {
      "url": "https://mcp.mythex.ai/mcp"
    }
  }
}

Toggle it on, click Connect when Cursor prompts, and approve in the browser. There are no API keys to paste: it's OAuth with refresh tokens.

A first prompt

Create a new Mythex project called "booking-demo": a React + TypeScript app where clients book 30-minute slots. Store bookings in Postgres, start the preview, check it in the browser and send me a screenshot. When it works, publish it.

The agent creates the project, writes the code in the sandbox, adds the database, starts the preview, checks it with a browser run, and publishes. You get back a preview URL, screenshots and a public URL.

Costs, honestly

• Connecting is free. Tool calls need a Mythex Pro account ($25/month), and sandbox time uses Mythex credits.
• Your model usage is billed by Claude or Cursor, not by Mythex.
• If the account isn't Pro or has no credits, the tool returns an upgrade_url instead of running.

Why do it this way

• No local setup. The app runs in a cloud sandbox, not on your laptop.
• One place for the boring parts. Database, storage, secrets, checkpoints and hosting stay in one project.
• You keep the code. It's ordinary React/TypeScript you can export or push to GitHub.

If you'd rather build in a chat UI, the same projects open in the Mythex web app. It also works in Arabic and 68 other languages.

Docs: https://docs.mythex.ai/features/mcp