Sometime ago, I created an MCP server with C#. It was just a basic project I was trying to learn how MCP works, and I followed this Microsoft blog
to get started. At that time, MCP was booming, and I wanted to explore AI development using my favorite language, C#. So, I decided to create a simple Google Sheets MCP integration just to see how it works.
This blog is not a step-by-step tutorial on how to create an MCP server, but more about what I think of MCP after building one.
So, MCP or Model Context Protocol is an open-source protocol that allows you to expose your methods and functions to AI agents. Basically, it lets the AI directly access your code or functions. That’s why developers get excited about it: it feels like connecting your codebase directly to an intelligent assistant.
Now, about why I don’t really love MCP. I don’t hate it — it’s just “okay” for me. AI is powerful, but it’s still not fully accurate or secure. If you’ve worked with AI before, you know how easily it gets confused or stuck repeating the same thing. MCP depends heavily on natural language, which makes it even more unreliable in real use cases.
For example, let’s say you build a simple MCP tool that just adds two numbers. When you connect your AI agent and ask, “Can you add 1 + 1?”, there’s a high chance the AI will just respond with “2” directly instead of calling your MCP tool — unless you explicitly tell it to use the tool. That’s the issue: MCP sounds great in theory but doesn’t always behave as expected in real workflows. It’s more suited for AI coding experiments or vibe coding sessions rather than serious projects where you’re writing the logic yourself.
Another thing I noticed while experimenting is that MCP feels more like a bridge technology something that fills the gap between AI tools and real-world codebases, but not something most developers actually need daily. It looks impressive in demos, but when you implement it, you realize it doesn’t really solve a new problem; it just repackages an old one with a different label.
I also think the hype around MCP mostly comes from how Big Tech markets it rather than how developers use it. Everyone loves to say “AI agents can now run your code,” but in practice, 90% of the time it’s faster and safer to just expose an API or run a script yourself. MCP adds another layer that looks fancy but doesn’t necessarily improve your workflow. There are also more developers building MCP tools than actual users. It’s mostly used by vibe coders or enthusiasts rather than everyday developers writing real-world code. People love the idea of AI and anything with “AI” in it gets instant hype but that doesn’t mean it’s always practical.
That said, I still respect the idea behind MCP. It’s a good step toward standardizing how AI tools interact with software, and I can see it becoming useful once AI models get more consistent and reliable. Maybe in the future, MCP will make more sense but for now, it feels more like a presentation feature than a production feature.
Top comments (0)