If you're building AI applications with .NET or constantly learning, you've noticed LLMs confidently give you code that doesn't compile or completely wrong explanations about how things work.
I got tired of it, so I built DotNet AI MCP Server.
**
What it does:
**
Connects your favorite client to two sources:
Live GitHub repos - Semantic Kernel, OpenAI .NET SDK, MCP C# SDK, AutoGen, and more. Real code and documentation from the actual repos.
Microsoft Learn - I proxied the official Microsoft Learn MCP tools but optimized them: better token efficiency, clearer descriptions, and improved argument names so the LLM actually picks the right tool.
**
The key difference:
**
Zero prompt engineering. Just ask your question naturally - "How do Semantic Kernel agents work?" or "Show me how to build an MCP server with C#" - and the tools trigger automatically. No need to tell it "use this tool" or "search the documentation" like other MCP servers require.
It uses progressive file exposure (repos → folders → files → content) which saves tokens and doesn't flood your context with irrelevant data.
**
Currently tracking:
**
AI Frameworks & LLM SDKs: Semantic Kernel • AutoGen • Kernel Memory • OpenAI .NET • Google Gemini • Anthropic Claude • MCP C# SDK • LangChain.NET • OllamaSharp
Vector Database C# SDKs: Pinecone • Qdrant • Weaviate • Redis Stack
Setup takes 30 seconds. If it helps, drop a ⭐ so other .NET devs can find it.
Try it: https://github.com/Ahod26/dotnet-ai-mcp-server
Roast me if it sucks. 🔥
Top comments (0)