azure-container-apps-ai-mcp

This project showcases how to use the MCP protocol with OpenAI, Azure OpenAI and GitHub Models. It provides a simple demo terminal application that interacts with a TODO list Agent. The agent has access to a set of tools provided by the MCP server.

The current implementation consists of three main components:

1. MCP Host: The main application that interacts with the MCP server and the LLM provider. The host instanciates an LLM provider and provides a terminal interface for the user to interact with the agent.
2. MCP Client: The client that communicates with the MCP server using the MCP protocol. The application providers two MCP clients for both HTTP and SSE (Server-Sent Events) protocols.
3. MCP Server: The server that implements the MCP protocol and communicates with the Postgres database. The application provides two MCP server implementations: one using HTTP and the other using SSE (Server-Sent Events).
4. LLM Provider: The language model provider (e.g., OpenAI, Azure OpenAI, GitHub Models) that generates responses based on the input from the MCP host.
5. Postgres: A database used to store the state of the agent and the tools.
6. Tools: A set of tools that the agent can use to perform actions, such as adding or listing items in a shopping list.

flowchart TD
    user(("fa:fa-users User"))
    host["VS Code, Copilot, LlamaIndex, Langchain..."]
    client[MCP SSE Client]
    clientHttp[MCP HTTP Client]
    server([MCP SSE Server])
    serverHttp([MCP HTTP Server])
    agent[Agent]
    AzureOpenAI([Azure OpenAI])
    GitHub([GitHub Models])
    OpenAI([OpenAI])
    
    tools["fa:fa-wrench Tools"]
    db[(Postgres DB)]

    user --> hostGroup 
    subgraph hostGroup["MCP Host"]
        host -.- client & clientHttp & agent
    end
    
    agent -.- AzureOpenAI & GitHub & OpenAI
    
    client a@ ---> |"Server Sent Events"| server
    clientHttp aa@ ---> |"Streamable HTTP"| serverHttp

    subgraph container["ACA Container (*)"]
      server -.- tools
      serverHttp -.- tools
      tools -.- add_todo 
      tools -.- list_todos
      tools -.- complete_todo
      tools -.- delete_todo
    end

    add_todo b@ --> db
    list_todos c@--> db
    complete_todo d@ --> db
    delete_todo e@ --> db
    
    %% styles

    classDef animate stroke-dasharray: 9,5,stroke-dashoffset: 900,animation: dash 25s linear infinite;
    classDef highlight fill:#9B77E8,color:#fff,stroke:#5EB4D8,stroke-width:2px
    
    class a animate
    class aa animate
    class b animate
    class c animate
    class d animate
    class e animate

    class container highlight

This demo application provides two MCP server implementations: one using HTTP and the other using SSE (Server-Sent Events). The MCP host can connect to both servers, allowing you to choose the one that best fits your needs.

To get started with this project, follow the steps below:

• Node.js and npm (version 22 or higher)
• Docker (recommended for running the MCP servers, and Postgres in Docker)
• An OpenAI compatible endpoint:
  • An OpenAI API key
  • Or, a GitHub token, if you want to use the GitHub models: https://gh.io/models
  • Or, if you are using Azure OpenAI, you need to have an Azure OpenAI resource and the corresponding endpoint.

1. Clone the repository.
2. Install the dependencies:

npm install --prefix mcp-host
npm install --prefix mcp-server-http
npm install --prefix mcp-server-sse

This sample supports the follwowing LLM providers:

In order to use Keyless authentication, using Azure Managed Identity, you need to provide the AZURE_OPENAI_ENDPOINT environment variable in the .env file:

AZURE_OPENAI_ENDPOINT="https://<ai-foundry-openai-project>.openai.azure.com"
MODEL="gpt-4.1"

# (optional) Set the Azure OpenAI API key if you are not using Managed Identity
# AZURE_OPENAI_API_KEY=your_azure_openai_api_key

And make sure to using the Azure CLI to log in to your Azure account and follow the instructions to selection your subscription:

az login

To use the OpenAI API, you need to set your OPENAI_API_KEY key in the .env file:

OPENAI_API_KEY=your_openai_api_key
MODEL="gpt-4.1"

To use the GitHub models, you need to set your GITHUB_TOKEN in the .env file:

GITHUB_TOKEN=your_github_token
MODEL="openai/gpt-4.1"

This project includes a DevContainer configuration that allows you to run the MCP servers in a containerized environment. This is the recommended way to run the MCP servers, as it ensures that all dependencies are installed and configured correctly.

Once you have opened the project in a DevContainer, you can run the MCP servers using the following the Docker section below.

You can run both MCP servers in Docker containers using the provided Docker Compose file. This is useful for testing and development purposes. To do this, follow these steps:

1. Make sure you have Docker installed on your machine. Type docker compose in your terminal to check if Docker Compose is installed.
2. Navigate to the root directory of the project and run the following command to build and start the containers:

docker compose up -d --build

This command will build and start the HTTP and SSE MCP servers, as well as the Postgres database container.

3. Access the MCP host terminal by running the following command in a separate terminal:

docker exec -it mcp-host bash

4. Inside the container, you can run the MCP host and interact with the LLM agent as described in the Usage section above.

1. First, run the MCP servers, in separate terminals:

npm start --prefix mcp-server-http
npm start --prefix mcp-server-sse

1. Run the MCP host in a separate terminal:

npm start --prefix mcp-host

You should be able to use the MCP host to interat with the LLM agent. Try asking question about adding or listing items in a shopping list. The host will then try to fetch and call tools from the MCP servers.

You can use the DEBUG environment variable to enable verbose logging for the MCP host:

DEBUG=mcp:* npm start --prefix mcp-host

Debugging is enabled by default for both MCP servers.

This project is licensed under the MIT License. See the LICENSE file for details.