American Cloud MCP server: manage your cloud from Claude or Cursor

American Cloud has an official MCP server, open source under Apache-2.0. It connects Claude, Cursor, Codex, Windsurf, VS Code or any other MCP client to your account, so you can manage VMs, DNS, block and object storage, networking, managed Kubernetes, databases and WordPress in plain English.

It speaks the Model Context Protocol, the open standard these assistants use to call outside tools. Ask a question, get an answer. Give an instruction, watch it happen.

You: What's running in my account right now, and what does each VM cost per month?

You: Audit my firewall rules — is anything open to the whole internet that shouldn't be?

You: Spin up an Ubuntu VM in us-west with my SSH key and open ports 22, 80, and 443.

No scripts. No console tabs. No API documentation open in another window. One config snippet and your AI assistant becomes a capable cloud operator.

What can the American Cloud MCP server manage?

The server ships more than 170 tools in seven service groups, covering the American Cloud platform:

Compute, storage, networking and Kubernetes are on by default. Databases, WordPress and DNS load with the --services flag, so you only carry the tools you use. And cost-estimate tools are included throughout, so your assistant can answer "what would a 3-node Kubernetes cluster cost?" before anything gets created.

What can the assistant change?

We know what you're thinking: an AI with access to my production infrastructure?

That's why the server is designed around a simple principle — read-only by default, writes by explicit opt-in:

• Out of the box, the assistant can look but not touch. Only read tools — list, inspect, cost estimates — are available until you explicitly enable management with a flag.
• Read-only API keys make it a guarantee. Provision a read-only key and writes are impossible at the account level, no matter how the server is configured. Give your AI a read-only key and let it audit everything — it physically cannot break anything.
• Destructive actions are flagged. Deletes and other irreversible operations carry a destructive marker, so MCP clients that support confirmations will ask you first.

Where does the server run?

The server runs locally, on your machine. Your API keys go straight from your environment to the American Cloud API — they never transit a third-party service, and there's no hosted middleman sitting between your AI assistant and your cloud.

That's not an implementation detail; it's the same principle the rest of our platform is built on. Your infrastructure, your keys, your control. If you care about why that matters, read our take on data sovereignty in 2026.

It's also open source under Apache-2.0, and every release is published to npm with provenance attestation — cryptographic proof that the package you install was built from the public repository. Inspect it, audit it, fork it: github.com/American-Cloud/americancloud-mcp.

How do you connect Claude, Cursor or Codex?

Setup takes about a minute. Add this to your MCP client's configuration (Claude Desktop, Claude Code, Cursor, Codex, Windsurf, VS Code, anything that speaks MCP):

{
  "mcpServers": {
    "americancloud": {
      "command": "npx",
      "args": ["-y", "@americancloud/mcp"],
      "env": {
        "AMERICANCLOUD_API_CLIENT_ID": "your-client-id",
        "AMERICANCLOUD_API_CLIENT_SECRET": "your-client-secret"
      }
    }
  }
}

Create an API key (start with a read-only one) in the console, restart your client, and ask: "What regions can I deploy to?"

We've written step-by-step guides for each client:

• Overview and quick start
• Claude Desktop
• Claude Code
• Cursor
• Codex
• Windsurf, VS Code, and other clients
• Things to try

What people are doing with it

A few patterns that emerged immediately in our own use:

The infrastructure audit. Hook up a read-only key and ask the assistant to review everything: what's running, what it costs, which firewall rules are wider than they need to be, which volumes aren't attached to anything. Twenty minutes of conversation replaces an afternoon of clicking.

Incident triage. "Why can't I reach my app?" The assistant checks the VM's power state, walks the firewall rules, verifies port forwarding, and inspects load balancer membership — in one conversation, while you drink your coffee.

Cost-aware provisioning. Because cost estimates are tools too, the assistant can price out infrastructure before creating it: "Compare the monthly cost of one large VM versus three small ones behind a load balancer."

Build and deploy in one session. This is our favorite. In a coding agent like Claude Code, the assistant can write your application and stand up the infrastructure to run it — create the VM, configure the firewall, point DNS at it, then deploy over SSH. Code and cloud in the same conversation.

Part of a bigger story

The MCP server sits alongside our official TypeScript, Go, and Python SDKs and our Terraform provider, all built on the American Cloud Platform API. Whether your operator is a human with a terminal, a CI pipeline, or an AI assistant, American Cloud is built to be driven programmatically. For how other clouds' MCP servers compare on where they run and what they can change, see which cloud providers have an MCP server.

Get started

1. Create an API key — start with read-only
2. Follow the setup guide for your client
3. Ask your assistant: "What's running in my account?"

Found a rough edge or want a tool we don't have yet? Open an issue — the server is open source and we're actively expanding it.