antrieb-mcp-server

An MCP server that runs and validate your AI-generated infra code. On real VMs. E.g. tell Claude to generate a shell script to setup a LAMP stack on Ubuntu and Antrieb will spin up a VM, run the script, self-correct until it works, and finally destroys the VMs.

No containers. No sandboxes. Real VMs with full OS access, networking, and multi-node clusters.

Antrieb is a remote MCP server, thus nothing to install. Add it to your config and start deploying. Antrieb uses its own hypervisor to control the full VM lifecycle.

> No credentials or cloud account required. Antrieb runs entirely on its own infrastructure, including real AWS provisioning. The terraform-aws and cloudformation-aws images run against real AWS using Antrieb's own credentials. You don't need to supply yours.

Antrieb: Make AI-Generated Infrastructure Converge.

[](https://youtu.be/oB6CTjDceMI)

Add this to your MCP client config (.mcp.json, Claude Desktop settings, etc.):

{
  "mcpServers": {
    "antrieb": {
      "type": "http",
      "url": "https://antrieb.sh/mcp"
    }
  }
}

OR with an API key:

{
  "mcpServers": {
    "antrieb": {
      "type": "http",
      "url": "https://antrieb.sh/mcp",
      "headers": {
        "Authorization": "Bearer ant_YOUR_API_KEY"
      }
    }
  }
}

That's it. No local install, no dependencies, no Docker.

You can try without an API key or get one by logging in at https://antrieb.sh/

> Note: Without an API key, your tasks will appear in the community feed.

You describe what you want in natural language. Your LLM generates the code, Antrieb spins up real VMs, executes it, validates the result, and self-corrects if something fails — all in one call.

Write a bash script that sets up a Node.js Express hello world app with PM2 then validate