mcp-on-azure

This MCP server is a minimal Express.js application that interfaces with Azure services to read image data from blob storage and generate descriptive captions using Azure OpenAI. It exposes two tools via HTTP endpoints: list_captain_files and describe_blob. The list_captain_files tool fetches and returns the names of all blobs in the configured container, while describe_blob accepts a blobName and queries Azure OpenAI's gpt-4o vision model to produce a short, clear description of the image. To use, start the server and then call the endpoints (for example via curl or your MCP client) at http://localhost:3333/tools/list_captain_files and http://localhost:3333/tools/describe_blob. Ensure your Azure credentials and container name are configured in the environment so the server can access both Blob Storage and OpenAI resources.

Prerequisites:

• Node.js 18+ (recommended 18/20/22)
• Git
• Access to an Azure Storage account with a container (e.g., captain-azure) containing images
• Azure OpenAI resource with a gpt-4o deployment

Install steps:

1. Clone the repository

git clone https://github.com/OfekBenEliezer/mcp-on-azure.git
cd mcp-on-azure

2. Install dependencies

npm install

3. Create and configure environment variables

• Copy the example and edit as needed

cp .env.example .env

• Populate the following variables in .env (or set them in your environment):

AZURE_STORAGE_CONNECTION_STRING=DefaultEndpointsProtocol=https;AccountName=<name>;AccountKey=<key>;EndpointSuffix=core.windows.net
CONTAINER_NAME=captain-azure
AZURE_OPENAI_ENDPOINT=https://<your-openai>.openai.azure.com
AZURE_OPENAI_KEY=<your-openai-key>
AZURE_OPENAI_DEPLOYMENT_NAME=gpt-4o
AZURE_OPENAI_API_VERSION=2024-02-15-preview

4. Start the server

node server.js

5. Verify the server is running

# MCP Server running on: http://localhost:3333
curl -s http://localhost:3333/tools/list_captain_files -X POST -H "Content-Type: application/json"

Tips and notes:

• The server uses environment variables to connect to Azure Blob Storage and Azure OpenAI. Do not commit these keys.
• If your blobs are private, consider generating SAS URLs with short TTL or use Managed Identity where supported.
• The server listens by default on port 3333; adjust as needed in code if you want a different port.
• Ensure your Azure OpenAI deployment name (gpt-4o) and API version match what you’ve configured in Azure.
• For production, restrict CORS and ingress and consider adding authentication layers for the MCP endpoints.