.NET OpenAI MCP Agent

What is .NET OpenAI MCP Agent?

This is an MCP agent sample app written in .NET, using Azure OpenAI, that connects to a remote MCP to-do server written in TypeScript. The agent’s host and client are built with .NET Blazor, and both the client and server are deployed to Azure Container Apps. It is intended for developers who want to see a full MCP implementation with a Blazor UI and Azure OpenAI.

How to use .NET OpenAI MCP Agent?

Use GitHub Codespaces or clone the repository with azd init -t openai-mcp-agent-dotnet. Set the environment name, then deploy with azd up. After deployment, navigate to the client app URL, log in (or bypass login by setting USE_LOGIN false), and enter natural‑language prompts like “Give me list of to do.” or “Set meeting at 1pm.” Clean up with azd down --force --prune.

Key features of .NET OpenAI MCP Agent

• .NET Blazor front end for the MCP host/client.
• Uses Azure OpenAI with the GPT-5-mini model by default.
• Remote MCP server (TypeScript) for to‑do operations.
• Both apps run on Azure Container Apps.
• Built‑in ACA authentication protects the client app.
• Deployable from the command line with azd up.

Use cases of .NET OpenAI MCP Agent

• Manage a to‑do list via natural language commands.
• Prototype an MCP agent with a web UI and cloud deployment.
• Demonstrate MCP client–server separation and secure communication.
• Showcase .NET and Azure OpenAI integration with an MCP tool server.

FAQ from .NET OpenAI MCP Agent

What does this agent do versus other MCP implementations?

This agent provides a full C#/.NET MCP host and client that talks to a remote TypeScript MCP server. It is a sample to illustrate the MCP pattern using Azure OpenAI and Azure Container Apps.

What are the runtime dependencies?

You need the .NET 9 SDK, Visual Studio Code with C# Dev Kit, node.js LTS, Docker Desktop or Podman Desktop, and an Azure subscription with an Azure AI Foundry instance.

Where does the data live?

The to‑do data is managed by the remote MCP server, which is deployed in your own Azure Container Apps environment. No data is stored outside your Azure subscription.

Are there known limits?

The sample uses the GPT-5-mini model by default and is intended for demo/prototyping. In production, you may need to adjust the model, auth mode (keyless vs. API key), and login settings.

What transports and authentication are supported?

The MCP client and server communicate over HTTP/HTTPS. The client app is secured by Azure Container Apps built‑in authentication (can be toggled off). The server is only accessible from the client app. Azure OpenAI can use either keyless (Microsoft Entra) or API key authentication, controlled by the AOAI_USE_API_KEY environment variable.