Other languages vs. C#

[aronweiler]

Pre-submission Checklist

• I have verified that this discussion would not be more appropriate as an issue in a specific repository
• I have searched existing discussions to avoid duplicates

Discussion Topic

I'm looking for peoples' honest opinions- coming from a background of C#, Python, and a bunch of other languages- What should I be implementing MCP servers in for longevity?

I have a number of MCP servers. some Go, Python, and C#- but it seems to me like the Python support is so much better than any other language that it makes it hard to stay with the others, despite the obvious advantages. It feels like if I want to be on the frontier of anything AI related I end up having to be on Python.

I would like to be frank here (since C# is really my preferred language) and ask how many other C# folks are in the same boat, and find the ecosystem here less than ideal.

[ChristianWeyer]

This is a great question, and one I’m also contemplating.

What makes you think the Python version is so much better? I see quite a lot of activity in this repo as well.

Thanks.

[mikekistler]

Yes, please share! What features does the Python SDK provide that are missing or inferior in the C# SDK?

The C# SDK is in active development and we are tracking the spec changes closely. For example, the C# SDK had full support of the 2025-06-18 spec within a week of its release.

So what's missing?

[ChristianWeyer]

E.g., I am implementing authN with both SDKs, and the C# is so much ahead of the Python one.

[PederHP]

Convenient middleware, which was a gap compared to the Python SDK with FastMCP, has just been added.

I've written servers and clients in Python, TypeScript, and C#, and while I am obviously biased, I do genuinely think the C# SDK has the best developer experience.

Language preferences are what they are, but all other things being equal, I think the SDK is fully par or better than the two main ones (Python and TypeScript).

[aronweiler]

I much prefer C# over the other languages by a lot- but when it comes to tooling availability around all things AI, Python has everyone beat. Their tools come out first, have the most features, and are generally the most well-supported. As far as writing a server like MCP and actually serving content, or doing anything that requires enterprise-level code, type-safety, performance, etc., there is really no contest- it's clearly C#.

I suppose long-term C# will be the winner here, as it's going to ultimately be the more stable, better performing option, and python is just the wild-west. It's just frustrating to see the majority of the AI community focused on python.

Also for clarity @mikekistler and @ChristianWeyer, I was not talking specifically about the C# MCP SDK- I am talking generally about the language ecosystems, and the availability of tools and support- but that does factor greatly into the choice of framework/SDK here.

[PederHP]

For classic machine learning, you're right. Training models and such. But for general tooling around all things generative AI? Microsoft.Extensions.AI is really good. There is Anthropic.SDK, which is a great wrapper (also compatible with MEAI). You can use open source models easily. REST APIs integrate smoothly with C# and .NET for image generation models.

And that's pretty much all you need. (and if it isn't there is a lot more good stuff). The rest is software development, engineering, and architecture. Something .NET excels at. So no, I don't think Python has everyone beat. I've written MCP servers in Python and it's by no means a better experience. Sure, when a new API comes out for some LLM - the Python and Typescript SDKs are usually first, but the REST APIs are too. So, I think we've already begun to see what you outline as long-term - and I agree - it's frustrating to see how Python is the default, even when other languages would be just as good - whether that's C#, Rust, Go, or TypeScript. But it's changing. So don't be afraid to hop onboard and use C# right now if that's where your organization or your own skills are. It's not like classic ML and data science where you'll still have a bad time if you don't use Python. Generative AI has good polyglot support and excellent C# support.