Building an MCP Client

So far in this book, we have spent all of our time on the server side. We have built tools, exposed resources, defined prompts, configured transports, and wired up dependency injection. Every time we wanted to test our work, we reached for an existing client like Claude Desktop or GitHub Copilot in VS Code. That is perfectly fine for many scenarios, but what happens when you need to talk to an MCP server from your own code?