HPC/AI Cluster Tool FAQs

Question 1

What is the HPC/AI Cluster Tool?

Accepted Answer

It's a minimal Model Context Protocol (MCP) server specifically for Azure HPC/AI clusters. It queries Kubernetes to retrieve essential GPU node information and InfiniBand topology details.

Question 2

How does this tool gather information from Kubernetes?

Accepted Answer

The tool uses synchronous `subprocess` calls to execute `kubectl` commands, processing the raw output into JSON-serializable Python data structures for easy integration and analysis.

Question 3

What specific data can this tool retrieve from my cluster?

Accepted Answer

It can list GPU nodes with their names, labels, and Ready/NotReady status. Additionally, it retrieves InfiniBand-related topology labels such as agentpool, pkey, and torset for each node.

Question 4

Can I test or develop with this tool without a full MCP host?

Accepted Answer

Yes, the `invoke_local.py` helper script allows you to execute server tools locally and in-process for testing and development, eliminating the need for an MCP host.

Question 5

What are the main prerequisites for running this tool?

Accepted Answer

You need Python 3.10+ installed and `kubectl` configured with access to your Azure HPC/AI Kubernetes cluster.

HPC/AI Cluster Tool

Key Features

Use Cases

HPC/AI Cluster Tool

Key Features

Use Cases