In this post I want to share the results of my initial tests on the Microsoft Learn Docs MCP Server. Which has recently become publicly available.
To clarify, MCP stands for Model Context Protocol. Which provides a standard way to connect AI models to different sources and tools.
MCP Servers are focused services that connect to different sources and tools in a standardized way. Which can be hosted in various locations. For example, running locally within a container or cloud hosted.
In this post I show my initial tests from an MCP client. Which in this case is Visual Studio Code running GitHub Copilot chat in Agent Mode.
From there the client connects via the MCP Protocol to the Microsoft Learn Docs MCP Server. Which then connects to Microsoft Learn. As shown in the below diagram.
I decided to base my initial tests on the DP-600 and DP-700 Microsoft Fabric exams. However, you can do a lot more. Especially when you decide to work with multiple MCP servers.
Along the way I share plenty of links and highlight some interesting facts.
Initial tests on the Microsoft Learn Docs MCP Server
To get started I needed to add a reference to an MCP Server in Visual Studio Code. I decided to do this with the Command Palette in Visual Studio Code as below.
I then filled out the details to connect to the Microsoft Learn Docs MCP Server.
Once the MCP server was up and running I opened up the GitHub Copilot chat in agent mode. I then asked what are the differences between the DP-600 and the DP-700 Microsoft fabric exams according to the Microsoft Learn documentation as below.
I then decided to check what would happen if I had worked with GitHub Copilot chat in Ask mode instead. Which I showed in a previous post. In addition to not utilizing the MCP server there was a hallucination in the results. As you can see below.
Which I think highlights the fact that working with an MCP Server can get you better results.
Anyway, I then took it a step further and asked chat to create a comparison table that showed all the same skills measured for both the DP-600 and DP-700 Microsoft Fabric exams based on their study guides. I got the below results.
As you can see, the table appears to be reasonably accurate. Towards the bottom GitHub Copilot asked me to let it know if I wanted a downloadable table. So I asked GitHub Copilot chat to provide me a table in HTML format. I was very impressed with the result.
Final words
I hope that sharing initial tests on the Microsoft Learn Docs MCP Server inspires some of you to look into it more.
My tests in this post were basic. However, you can do a lot deeper testing to see if it fulfils your needs. For example, you can look to work with it to generate a study plan for a Microsoft exam. Like the DP-600 and DP-700 Microsoft Fabric exams mentioned in this post.
You can even take it one step further and connect GitHub Copilot chat to multiple MCP servers to perform deeper tasks based on multiple sources. For example, you can also add an Azure or a Microsoft Fabric MCP server.
Of course, if you have any comments or queries about this post feel free to reach out to me.
Be First to Comment