Amazon SageMaker AI MCP Server

uvx awslabs.sagemaker-ai-mcp-server@latest

{
  "mcpServers": {
    "sagemaker-ai-mcp": {
      "command": "uvx",
      "args": [
        "awslabs.sagemaker-ai-mcp-server@latest"
      ]
    }
  }
}

AWS ships this one inside the awslabs MCP monorepo, and its current scope is narrower than the name suggests: the tools cover SageMaker HyperPod, meaning cluster deployment, node management and lifecycle operations for clusters orchestrated by either Amazon EKS or Slurm. What makes it safe to hand to an agent is that capability is opt-in at launch. Mutating operations only exist when the process is started with --allow-write, and reading logs and cluster events requires --allow-sensitive-data-access, so the default configuration is a read-only view of your training fleet. Credentials come from the ordinary AWS CLI profile rather than a key in the client config, and the server runs through uvx on Python 3.10 or later.