Why we need Roslyn MCP?

[niltor]

Essential Argument for Roslyn MCP + Preinstalled .NET 🚀

1. AI coding for .NET cannot be reliable without Roslyn MCP

Current AI tools suffer from uncertainty, hallucination, instability, and missing context.
These issues are inherent when AI operates only on text.

Roslyn MCP provides compiler‑level truth—syntax trees, symbols, diagnostics, and full project structure—turning AI coding from guesswork into deterministic, verifiable engineering.

---

2. Roslyn MCP enables capabilities no text‑based AI can achieve

With Roslyn MCP, AI gains:

• Accurate solution structure instead of slow file crawling
• Analyzer errors, warnings, and suggestions, enabling AI to follow existing rules and architecture
• Instant semantic navigation, dramatically improving speed and correctness

This transforms AI from a “code generator” into a true semantic co‑developer.

---

3. Without Roslyn MCP, Visual Studio risks losing competitive ground ⚠️

Other ecosystems are already integrating AI with their semantic engines:

• JetBrains with PSI
• VS Code with MCP-based language services
• Third‑party tools moving toward compiler‑integrated AI

If Visual Studio does not move now, others will define the AI‑first IDE experience.

---

4. Preinstalling .NET is a strategic multiplier

Python’s rise was accelerated by being preinstalled on major operating systems.
Preinstalling .NET on Windows—and partnering with Linux distributions—would:

• Remove onboarding friction
• Expand .NET’s presence in scripting, automation, and education
• Strengthen the ecosystem long‑term

This is not about politics; it is about strategic reach and competitiveness.

---

Final Condensed Statement (Leadership‑Ready) 💡

Roslyn MCP is essential for trustworthy AI coding in .NET.
It provides the semantic foundation needed to eliminate hallucinations, improve reliability, and enable advanced AI‑driven refactoring and diagnostics.
If Microsoft does not deliver Roslyn MCP soon, competing tools will, and Visual Studio will lose strategic advantage.
At the same time, preinstalling .NET across Windows and partner Linux distributions would replicate Python’s success and significantly expand the .NET ecosystem.

---

[CyrusNajmabadi]

If you want, I can craft a one‑page executive brief, a GitHub proposal, or a public advocacy post based on this.

Please don't post ai slop.

--

Followup: @niltor removing the ai slop line doesn't make the PR not slop. It just tries to hide it. Please write an actual real analysis.

[niltor]

I don’t know what your problem is. Instead of focusing on the content itself, you’re fixated on whether AI was used. I use AI for translation and content polishing, and I don’t see anything wrong with that! Get back to the issue itself, focus on solving problems, not just creating arguments!

[333fred]

No mcp is necessary. All major llm harnesses support lsp natively at this point, and https://github.com/dotnet/skills/ provides a plugin for using it automatically. In the future, while we appreciate the enthusiasm, a giant wall of text that has very obviously been AI generated with very little effort into cleaning it up is not appreciated.

[jasonmalinowski]

(this is somewhat duplicative: @333fred posted while I was typing)

As far as the content is concerned, we are already well aware of the shortcomings of AIs today and the various places Roslyn could help address those shortcomings. We have a number of experiments and efforts going to try various things, including some of the things you're suggesting. This wasn't really telling us anything we didn't already know, which can make it feel pretty "low value".

And in regards to an "MCP" -- there's a bit more to the story of how we'd integrate anyways. For example, Copilot has increasing support for working with LSP servers, and we're working on registrations to that effect.

[CyrusNajmabadi]

I don’t know what your problem is. Instead of focusing on the content itself,

I was commenting on the content. It was unverified ai slop. This was not translation. It was generation of something an ai came up with that was not validated first.

Using ai is fine if done responsibly. Simply having it generate the whole thing and taking it verbatim to dump for others to read is not.