Experiment to use Microsoft Learn documentation, Microsoft Azure, and Microsoft Foundry without visiting their websites.
Hi
We are going to run an experiment to use Microsoft Learn documentation, Microsoft Azure, and Microsoft Foundry, without visiting their websites. First I would like you to create a markdown file that captures all my messages to you. (You can simply copy paste)
Great. Let's see what cli commands I need to create a Microsoft Foundry instance?
Ok, let's create another file that collects all the scripts we are going to run, which can later be replayed to repeat the experiment. The first command should be if the current user has az cli installed and if they are signed in
yes, run it
Yes, let's add the foundry creation script. Take it easy though: after each command verify that it worked, break if it failed
let's go
My goal is to validate the correct deployment of Azure resources through az cli using a small test client app. Before we create the app, can you confirm the Azure and Foundry resources are ready for a client to connect? Anything missing? A model perhaps (gpt-4.1-mini)?
ok. let's create a dotnet console app that connects to Learn MCP Server, lists MCP tools, and uses the docs search tool. Cheating allowed from a previous attempt https://raw.githubusercontent.com/pdebruin/learndocsmcpconsole/refs/heads/main/2af/Program.cs
yes go ahead and upgrade az cli
If this approach has a dependency on a certain version of az cli, I would expect that to be part of the script. Can you verify?
Right, so the script works and the test client too. Let's optimize for maintainability. First, take a look at the script. It is 215 lines. Is that really necessary for a repeatable foundry creation script?
I love simplification. This looks like a good time to add commit and push to github. The changes seem to affect 53 files. Looks like a missing .gitignore?
Let's see if the testclient can be simplified too: yours is 134 lines, mine was 42. Also look at silly outputs like console.writeline of constant strings. And there is some funky threading going on, whereas I prefer simplicity for maintenance and don't mind if the ui thread is blocked
when I run the app, there is no output. I want to see the response to the user message, or an error message. Can you verify?
This is just for me to learn: I my previous sample I used persistentAgentsClient.CreateAIAgentAsync and now you are doing threads and runs and cleanup. Why the change? And is there no threadless way?
Have you looked at Microsoft Agent Framework at all? Could that help?
Yes, use that
Ok, this time don't change anything, just self-reflect. I still see the threading going on in this code. However in a sample like https://github.com/microsoft/agent-framework/blob/main/dotnet/samples/GettingStarted/ModelContextProtocol/FoundryAgent_Hosted_MCP/Program.cs there is none of that. Seriously, why would you want to keep the threading?
Can you recreate the client based on the last link?
Better. Before calling docs_search, can you ask the server to list all the tools available and write those to the console?
Self-reflection time, don't change anything: You have made it work using low level http requests. What do you think when you compare it to something like:
let's do the elegant approach
let's add commit push to github