Build your own MCP server in .NET

Microsoft.Extensions.Hosting
This package provides the .NET Generic Host infrastructure. It gives your project a structured startup pipeline, dependency injection, logging, and lifecycle management. The MCP server runs as a hosted application, so this package forms the backbone of the server runtime.

OllamaSharp.ModelContextProtocol
This package contains a community implementation of the Model Context Protocol for .NET. It provides the attributes and server tooling used to expose methods as MCP tools, such as McpServerTool and McpServerToolType. Without this package, the server wouldn’t know how to register or publish tools to MCP-compatible clients.

What it does

Here’s a high-level summary of what the tool offers (for now):

• It is built in .NET (C#) and implements an MCP server to provide “time”-related functionality — for example retrieving current time in various formats, converting between time zones etc.
• The idea is: if an AI agent or application needs to reason about time — “What’s the time difference between Brussels and New York?”, “Schedule a job for 09:15 tomorrow in CET”, “Convert this timestamp to UTC” — instead of you writing bespoke logic each time, you provide a standardized server that supports these operations.
• The server exposes the tools via the MCP protocol, so a host (AI agent) can discover: “Okay, there’s a tool for time conversion”, call it with parameters, get results, all through the MCP mechanism.

Why I built it

• Time logic is deceptively tricky in distributed systems: time zones, daylight saving shifts, scheduling semantics, time arithmetic — these are classic sources of bugs. By exposing this logic via an MCP server, I make it accessible in a clean, reusable, discoverable fashion.
• It’s a testbed for MCP in .NET: building this server gave me hands-on experience with the MCP C# SDK (and protocol), tool discovery, message handling, transports.
• It enables AI-enabled workflows: Say you have an agent that needs to plan meetings, schedule tasks, generate reports with time stamps — now the agent can call this “time tool” server rather than hope the time logic is baked into the model or ad hoc. That makes your system more robust, maintainable.
• It sets a foundation for expansion: Today it’s time-tools, tomorrow you can imagine date-calendars, scheduling conflict detectors, aggregated logs by time period, etc., all exposed via the MCP pattern.

Securing your MCP Server

One thing to keep in mind when you start building your own MCP server — like my McpTimeTools — is security.

The Model Context Protocol (MCP) was designed to make tools easily discoverable and callable by AI agents. That’s great for interoperability, but it also means: if your server runs without any form of authentication, any client who can reach it could technically call your tools.

Here are the most useful best practices for security on an MCP server

• Always use TLS (HTTPS) to encrypt traffic.
• Require authentication (for example an API key, bearer token, or OAuth flow).
• Implement authorization to restrict which clients can call which tools.
• Apply the principle of least privilege — expose only what’s needed.
• Keep credentials and tokens out of source code, and rotate them regularly
• Use rate limiting to prevent abuse and control usage costs

How secure is an MCP server?

By default, the Model Context Protocol (MCP) doesn’t enforce any authentication or authorization. That means an MCP server is only as secure as the protections you add around it. In other words: security is the server’s responsibility.

(JSON Web Tokens) to control access. In C#, you can use the built-in ASP.NET Core authentication middleware to validate tokens issued by an authorization server such as Keycloak or Azure Entra ID. This ensures that only clients with valid tokens can access your tools or APIs.

For example, the server can require a Bearer token in each request and verify its issuer, audience, and signature before granting access. When configured this way, an MCP server inherits the same protection model as any modern web API.

Beyond basic token validation, several best practices strengthen your MCP server’s security:

• Always validate tokens against the correct issuer and audience.
• Enforce HTTPS in production — never send tokens over plain HTTP.
• Giving clients access only to the tools they need.
• Store secrets and cached tokens in secure, encrypted storage.
• Never log sensitive data such as authorization headers or tokens.

When implemented with JWT validation, HTTPS, and strict scope handling, an MCP server can be as secure as any enterprise-grade API. The protocol itself is flexible — it’s the server’s configuration and adherence to standard authorization frameworks that determine its real security.