MCP

CloudBooster ships an MCP (Model Context Protocol) server that exposes infrastructure tooling to AI assistants. With it, Claude (or any MCP-compatible assistant) can look up AWS best practices, propose component compositions, generate Architecture Decision Records, and render infrastructure diagrams — all without leaving your editor.

How it works

You connect the CloudBooster MCP server to your AI assistant's MCP configuration. The assistant can then call CloudBooster tools as part of its reasoning — for example, when you ask "what's the best way to add a managed Redis cache to my staging environment?", the assistant can call aws_best_practices_for and aws_compose to produce a concrete, reviewable answer.

The MCP server is read-only by default. It proposes changes; applying them still goes through the normal ChangeSet review flow.

Setup

Prerequisites

• Python 3.10+ — the MCP server runs as a Python process via uvx.
• uv — the Python package runner. Install it with curl -LsSf https://astral.sh/uv/install.sh | sh or see docs.astral.sh/uv.
• A CloudBooster account — needed for composition endpoints (aws_composition_for, aws_propose_components, aws_resolve_composition). Catalog lookups (aws_lookup_primitive, aws_best_practices_for) work without one.
• A CloudBooster API key — generated from your account settings at cloudbooster.io. Required for authenticated endpoints.

Install

No separate install step is needed. The server runs on demand via uvx, which fetches the latest published version automatically:

uvx cbx-mcp@latest --help

If the command prints usage output, the server is correctly reachable.

Configure

Add cbx-mcp to your MCP client's server list. The examples below use the CB_API_KEY environment variable — the server also reads CB_ORG and CB_PROJECT if you want to scope requests to a specific organisation or project.

Claude Code (~/.config/claude-code/mcp_servers.json):

{
  "mcpServers": {
    "cbx-mcp": {
      "command": "uvx",
      "args": ["cbx-mcp@latest"],
      "env": {
        "CB_API_KEY": "your-api-key-here"
      }
    }
  }
}

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "cbx-mcp": {
      "command": "uvx",
      "args": ["cbx-mcp@latest"],
      "env": {
        "CB_API_KEY": "your-api-key-here"
      }
    }
  }
}

After saving the config, restart your MCP client. Open the MCP server panel — you should see the eight CloudBooster tools listed.

Optional environment variables:

Verify

Once the server is running, send this prompt to your AI assistant:

What's the best primitive for serving a static site on AWS?

The assistant should call aws_best_practices_for or aws_lookup_primitive, and the answer should mention CloudFront, S3, and ACM as the core primitives. If you see those three services, setup is complete.

Tools

Related

• CLI — run cbx audit aws from the terminal.
• API — the CloudBooster REST API.