Antrieb MCP Server

An MCP server that validates your AI-generated infra code. On real VMs. Tell Claude (or any MCP client) to generate the code to setup a LAMP stack and Antrieb will spin up VMs, run the code, and self-correct until it works, then 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.

Quick Start

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.

What It Does

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 it on Antrieb with 1 Ubuntu node"
"Write a Terraform stack that creates 2 t3.micro EC2 instances behind a Network Load Balancer and validate it on Antrieb"
"Write an Ansible playbook to set up Prometheus and node_exporter on 3 nodes then validate on Antrieb"

Every execution returns a live monitoring dashboard so you can watch it happen in real time.

Tools

Antrieb exposes 5 MCP tools:

`run`

Validate infrastructure code on real VMs.

Parameter	Type	Required	Description
`prompt`	string	yes	The code and intent to validate
`cluster`	array	no	VM topology (e.g. `["ubuntu24.04 x3"]`, `["ansible-controller", "ubuntu24.04 x3"]`)
`language`	string	no	`bash`, `python`, `ansible`, `dockerfile`, `terraform-aws`, `cloudformation-aws`
`session_id`	string	no	Resume a previous session for iterative changes

`search`

Find available VM images.

Parameter	Type	Required	Description
`keywords`	string	no	Search by name, description, or tags

`status`

Monitor a running job. Supports long-polling.

Parameter	Type	Required	Description
`job_id`	string	no	Job ID from a previous `run`
`timeout`	number	no	Long-poll timeout in ms

`files`

Download files from a completed session.

Parameter	Type	Required	Description
`session_id`	string	yes	Session ID
`filenames`	array	yes	Files to retrieve

`cancel`

Stop a running job and destroy its VMs.

Parameter	Type	Required	Description
`job_id`	string	yes	Job to cancel

Available Environments

Base Images

ubuntu24.04
almalinux9
archlinux
centos-stream10
alpine

Pre-Built Images

Antrieb includes specialized images with pre-configured environments:

ansible-controller, terraform-aws, cloudformation-aws, podman-docker

Use search to discover all available images, or just describe what you need and Antrieb will pick the right topology.

How It Works

Your LLM generates the automation code
You call run with the code or a natural language prompt
Antrieb provisions VMs — each VM in less than a second
Code executes on the real VMs
If something fails, Antrieb reads the error, rewrites the code, and retries
You get back the result, generated files, and a monitoring dashboard
Antrieb destroys the VMs

Monitoring

Antrieb monitoring dashboard

Every run returns a telemetry_url pointing to a Grafana dashboard with:

Real-time execution timeline
Generated files with syntax highlighting
Per-node output logs
LLM-generated digest of what happened

Multi-Node Clusters

Specify topology with the cluster parameter:

// Shell or Python on 3 identical Ubuntu VMs
{ "cluster": ["ubuntu24.04 x3"] }

// Ansible controller + 3 managed nodes
{ "cluster": ["ansible-controller", "ubuntu24.04 x3"] }

// Terraform controller against real AWS
{ "cluster": ["terraform-aws"] }

// Docker or Podman
{ "cluster": ["podman-docker"] }

// Shell or Python on 3 identical Alpine VMs
{ "cluster": ["alpine x3"] }

License

Apache 2.0

antrieb