🧠 Azure Self Help MCP Server

Welcome to the Azure Troubleshooter MCP Server – a hands-on implementation of Model Context Protocol (MCP), designed to enable intelligent agents (like Claude or Semantic Kernel) to troubleshoot Azure resources step-by-step using the Azure Self Help API.

In short: it lets your AI app talk to Azure’s Guided Troubleshooter, automating common diagnostic workflows.

💡 What Is This?

This project implements an MCP Server that exposes Azure's Self Help API as a toolset. Once connected to an MCP Host (like Claude Desktop or an SK App), you can ask the agent to:

• Create a troubleshooter session
• View its current step
• Continue the flow with your response
• End or restart the session

Think of this as your interactive AI assistant for Azure diagnostics.

✨ Live Demo Use Case

❓ “I can’t SSH into my Azure VM.”

This tool lets an LLM walk you through possible causes and fixes interactively using Microsoft’s Guided Troubleshooter API.

🧰 Prerequisites

• .NET 8 SDK
• Azure CLI (logged in)
• Claude App or any MCP Host
• IDE / Terminal

📦 Setup

Step 1: Install Packages

dotnet add package ModelContextProtocol --prerelease
dotnet add package Microsoft.Extensions.Hosting
dotnet add package Azure.ResourceManager.SelfHelp
dotnet add package Azure.Identity

Step 2: Create Your MCP Server

In Program.cs:

var builder = Host.CreateEmptyApplicationBuilder(settings: null);

builder.Services.AddMcpServer()
    .WithStdioServerTransport()
    .WithToolsFromAssembly(); // Registers [McpServerTool] methods

var app = builder.Build();
await app.RunAsync();

Step 3: Write Your MCP Tools

In AzureTroubleshooterTools.cs:

[McpServerToolType]
public static class AzureTroubleshooterTools
{
    [McpServerTool, Description("Create a troubleshooter session")]
    public static async Task<string> CreateTroubleshooter([Description("Resource Uri of the azure resource")]string scope)
    {
        string solutionId = "e104dbdf-9e14-4c9f-bc78-21ac90382231"; // this is solutionId for vm ssh issue. This id can be found using Discovery API of Azure Help, which will also be part of this MCP Server
        string troubleshooterName = Guid.NewGuid().ToString();
        var client = new ArmClient(new DefaultAzureCredential());

        var troubleshooterId = SelfHelpTroubleshooterResource.CreateResourceIdentifier(scope, troubleshooterName);
        var troubleshooter = client.GetSelfHelpTroubleshooterResource(troubleshooterId);

        var data = new SelfHelpTroubleshooterData
        {
            SolutionId = solutionId,
            Parameters = { ["ResourceURI"] = scope }
        };

        ArmOperation<SelfHelpTroubleshooterResource> lro = await troubleshooter.UpdateAsync(WaitUntil.Completed, data);
        return $"Troubleshooter created with ID: {lro.Value.Data.Id}";
    }

    // Tools: Get, Continue, End, Restart...
}

🤖 MCP Config for Claude or Other Hosts

To connect your MCP Server with an MCP-compatible app like Claude, update your config file (in Claude Desktop: Settings > Developer):

{
  "mcpServers": {
    "azurehelpTroubleshooter": {
      "command": "dotnet",
      "args": [
        "run",
        "--project",
        "/Users/yourname/Projects/AzureTroubleshooterServer",
        "--no-build"
      ]
    }
  }
}

Once saved and restarted, Claude will detect your tools — ready to assist with Azure diagnostics! ✅

🧪 Available Tools

🛡 Authentication

Uses DefaultAzureCredential, so it works out-of-the-box with:

• Azure CLI
• Environment variables
• Managed Identity

🧠 Why This Matters

Model Context Protocol (MCP) is revolutionizing how LLMs communicate with tools. By building an MCP Server for Azure Help APIs, you can:

• Make Azure debugging agent-native
• Empower AI apps with cloud intelligence
• Extend to other Microsoft APIs with the same structure

📎 Related Blogs

• Build an AI App That Can Browse the Internet Using Playwright MCP Server
• Step-by-Step Guide to MCP Servers

🙌 Acknowledgements

Thanks to the amazing Azure SDK team and the creators of MCP for making this kind of developer magic possible. ✨

• https://learn.microsoft.com/en-us/rest/api/help/
• https://docs.anthropic.com/en/docs/agents-and-tools/mcp