MCP-Serverless-Core

MCP-Serverless-Core is a .NET isolated Azure Functions project that exposes an MCP tool endpoint.
It includes a sample MCP tool (FibonacciTool) that generates Fibonacci numbers and can be used as a starter template for building custom MCP tools.

Project Brief

This project demonstrates how to host Model Context Protocol (MCP) tools on Azure Functions using:

• Azure Functions v4 (dotnet-isolated)
• .NET 10 target framework (net10.0)
• Microsoft.Azure.Functions.Worker.Extensions.Mcp
• Application Insights integration

The sample tool accepts a count parameter and returns a Fibonacci sequence.

Prerequisites

• .NET SDK (compatible with net10.0)
• Azure Functions Core Tools v4
  Install guide: https://go.microsoft.com/fwlink/?linkid=2174087
• Docker (optional, for local Azurite storage emulator)

Configuration

1. Create local.settings.json at the project root.
2. Use the example file as a baseline:

{
	"IsEncrypted": false,
	"Values": {
		"AzureWebJobsStorage": "UseDevelopmentStorage=true",
		"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated"
	}
}

Notes:

• AzureWebJobsStorage defaults to local emulator mode.
• Update connection values as needed for your environment.

Run Locally

If you want to use local storage emulation, start Azurite first:

docker compose up -d

Then run the Azure Function app:

func start

Build

dotnet build

Project Structure

• Program.cs: Function host startup and dependency registration.
• Tools/FibonacciTool.cs: Sample MCP tool implementation.
• host.json: Host-level settings, including MCP extension configuration.
• local.settings.json: Local app settings and storage configuration.

Sample Tool Behavior

Tool name: FibonacciTool
Description: Generates Fibonacci numbers up to the specified count.

Example response:

• Input count = 5 -> 0, 1, 1, 2, 3
• Input count = 1 -> 0

Notes

• MCP server metadata and behavior can be tuned in host.json under extensions.mcp.
• Keep secrets and environment-specific settings in local.settings.json (do not commit secrets).