Managing Servers with Words: How Red Hat Taught AI to Understand Admin Tasks

System administrators are used to working with commands, configuration files, and interfaces that require precise syntax. A single wrong character leads to an error. Want to find out which servers are acting strangely? You have to run the right commands, filter the output, and manually correlate the data. It works, but it's time-consuming and requires experience.

Red Hat has proposed a different approach: what if you could simply talk to your infrastructure?

What Is Red Hat Satellite and Why Does It Need AI?

Red Hat Satellite is a tool for centralized management of Red Hat Enterprise Linux-based servers. It helps monitor system health, manage updates, track configurations, and identify issues. When a company has tens or hundreds of servers, it's hard to get by without such a tool.

Now, an MCP server has been added to Satellite. Simply put, it's a software “translator” between a language model and the server management system. An administrator writes a query in natural language – for example, “show me hosts with outdated packages” or “which systems haven't been in contact for the last three days” – and receives a response generated from real data from Satellite.

No special syntax. No nested filters. Just a question – and an answer.

How It Works – Without Diving into the Details

MCP stands for Model Context Protocol – a protocol that allows a language model to “connect” to external tools and retrieve current data from them. The idea is that a language model on its own doesn't know what's happening in your infrastructure right now. But through an MCP server, it can request the necessary information in real time and provide meaningful answers based on it.

This setup uses two components: Goose CLI, a client for interacting with the language model via text, and Ollama, a tool for running language models locally on your own hardware. Together, they create an environment where an administrator can have a dialogue with the model, which in turn queries Satellite for data via the MCP server.

An important point: all of this can run locally, without sending any data to the cloud. This is crucial for corporate environments where the confidentiality of the infrastructure is critical.

What Can You Actually Do with This Tool?

Here are a few examples of tasks that can now be handled with a simple text query:

• Find hosts with pending security updates.
• Identify systems that haven't synchronized with Satellite in a long time.
• Get a status summary for a specific group of servers.
• Investigate the cause of a particular host's anomalous behavior.

This isn't a replacement for in-depth diagnostics, but it's a very convenient way to quickly get your bearings, especially when you need to grasp the big picture or narrow down the list of suspect systems before diving deeper into an issue.

For an experienced administrator, it's a time-saver. For those just getting started with Satellite, it's a way to get the data they need without knowing all the intricacies of the interface.

Why This Matters Beyond This Specific Product

As a protocol, MCP is rapidly gaining popularity in the AI tools ecosystem. More and more systems are getting their own MCP servers, and this is gradually changing how people interact with enterprise software.

Previously, language models were good at reasoning but poor at knowing the current state of external systems. MCP bridges this gap: the model gains access to “live” data and can provide answers that are grounded in reality, not just generate text.

In the context of infrastructure management, this is especially valuable because what matters isn't the model's abstract “intelligence,” but the accuracy and timeliness of the information it works with.

What Questions Remain Open?

For now, this is more of an early, experimental integration than a turnkey enterprise solution. Getting it running requires manually configuring several components – a task manageable for a technical specialist, but not something you can deploy in five minutes.

The question of reliability also remains open: how accurately does the language model interpret queries in specific scenarios, how does it handle non-standard phrasing, which tasks can it solve confidently, and where is it better to double-check the results manually?

Nevertheless, the direction is clear: managing infrastructure via natural language is no longer science fiction or a distant future. It's a working prototype that you can deploy and try right now.