Building MCP Servers
UMCP
1 days
Upcoming Sessions
Need a private training for your team? Request a private training
Not ready to book yet? Request an offer
Introduction to the Model Context Protocol
Why AI applications need a standard way to reach tools and data, and how MCP's architecture of hosts, clients and servers provides it.
- The integration problem MCP solves
- Hosts, clients and servers
- Tools, resources and prompts
- Transports: stdio and Streamable HTTP
- The MCP ecosystem: SDKs, clients and registries
- LAB: Use existing MCP servers from an AI client
Building an MCP Server
Build a server with the official C# SDK that exposes tools, resources and prompts, and test it before any AI client touches it.
- Setting up a server project
- Exposing tools with typed inputs and structured output
- Designing tools a model uses well
- Resources and resource templates
- Prompt templates
- Testing and debugging with the MCP Inspector
- LAB: Build an MCP server for an existing API
Advanced Server Features
Let a server ask for more input while handling a call, report progress and hand off long-running or interactive work.
- Multi round-trip requests
- Elicitation: form mode and URL mode
- Progress and cancellation
- Long-running operations with the Tasks extension
- Interactive UI with MCP Apps
Remote Servers and Security
Move a server from your machine to the cloud, and protect it and its users with OAuth 2.1 and Microsoft Entra ID.
- Stateless servers over Streamable HTTP
- Authorization with OAuth 2.1
- Protecting a server with Microsoft Entra ID
- Prompt injection, tool poisoning and other threats
- Deploying to Azure
- LAB: Secure a remote MCP server with Microsoft Entra ID and deploy it to Azure
Using MCP in Your Own Agents
Use MCP servers from your own code, so your agents can use the same tools as Claude or Copilot.
- Connecting to MCP servers from an MCP client
- Giving an agent MCP tools
The Model Context Protocol (MCP) is the open standard that lets AI applications such as Claude, ChatGPT and GitHub Copilot connect to tools and data in one uniform way. In this course you build MCP servers that expose your own APIs, data and workflows, and you connect them to existing AI clients and to your own agents.
With MCP you write one server that every MCP client can use, instead of a custom integration for each model and app. After this day you can design, build, secure and deploy an MCP server. You also know when MCP is the right choice and which risks it brings.
You have experience writing C# and preferably have built web APIs with ASP.NET Core. Experience with Azure or an LLM API helps but is not required.
