Building a Serverless MCP Tool Host with .NET 10 and Azure Functions

Ajran Hossain | Apr 17, 2026 min read
Building a Serverless MCP Tool Host with .NET 10 and Azure Functions

MCP-Serverless-Core is a minimal, production-ready starting point for hosting Model Context Protocol (MCP) tools on Azure Functions using the .NET isolated worker model.

The goal is simple: Expose MCP-compatible tools through a serverless endpoint that can scale to zero, while staying lightweight and extensible.

Why Serverless MCP?

Running MCP tools in a serverless environment provides:

  • Zero idle cost (scale-to-zero)
  • Automatic scaling
  • Minimal infrastructure overhead
  • Native integration with Azure monitoring (Application Insights)

This makes it ideal for AI agents, tool execution backends, and experimental workflows.

Tech Stack

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

Architecture

The project follows a clean and minimal structure:

/Program.cs              -> Host bootstrap & DI
/Tools/FibonacciTool.cs  -> Sample MCP tool
/host.json               -> MCP + host configuration
/local.settings.json     -> Local environment config

MCP tools are registered and exposed through the Azure Functions runtime, allowing external systems (like AI agents) to invoke them via a standardized interface.

Sample Tool: Fibonacci

A simple tool is included to demonstrate MCP integration.

Tool Name: FibonacciTool Input: count Output: Fibonacci sequence

Example:

count = 5
→ 0, 1, 1, 2, 3

This acts as a baseline for building more advanced tools.

Running Locally

Start storage emulator (optional):

docker compose up -d

Run the function app:

func start

Build:

dotnet build

Configuration

Create a local.settings.json:

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

MCP Configuration

MCP behavior can be tuned inside:

host.json → extensions.mcp

This allows customization of:

  • Tool exposure
  • Metadata
  • Execution behavior

Development Notes

  • Keep secrets out of source control
  • Use local.settings.json for local development
  • Extend the /Tools directory to add new MCP tools
  • Designed to plug directly into AI agent ecosystems

Screenshot

Add your VS Code running server screenshot here

MCP Server Running

Final Thoughts

This project is intentionally minimal but powerful.

It gives you:

  • A working MCP server
  • A serverless execution model
  • A clean base to build custom tools

From here, you can evolve it into:

  • AI agent tool backends
  • Internal automation services
  • Distributed tool execution systems

Source Code

Available on GitHub MCP-Serverless-Core (as always). Clone, extend, and build your own MCP-powered services.