.NETRoslynMCPAIopen source

Giving AI Agents Eyes Into Your .NET Codebase

When I started using AI coding assistants seriously, I kept running into the same wall: the AI would confidently suggest a refactor, get the type hierarchy wrong, and send me chasing a bug it introduced. Not because it was bad at coding — but because it was reading my codebase the way a text editor from 1995 would: as flat files, searched by pattern.

That’s why I built Roslyn CodeLens MCP.

What It Does

Roslyn CodeLens MCP is a Model Context Protocol server that gives AI agents real semantic understanding of .NET codebases. Instead of grep, it uses the Roslyn compiler API — the same engine that powers Visual Studio’s IntelliSense, refactoring, and diagnostics.

That means an AI agent can:

  • find_implementations — see every class implementing an interface, not just ones with obvious names
  • find_callers — know every call site before suggesting a rename
  • get_type_hierarchy — walk base classes and derived types correctly
  • get_di_registrations — understand the DI graph without running the app
  • get_diagnostics — read compiler errors and Roslyn analyzer warnings
  • apply_code_action — execute Roslyn refactorings with a diff preview before writing to disk

Over 30 tools in total, from dead code detection to cyclomatic complexity analysis.

Why MCP?

MCP lets any supporting AI client — Claude Code, VS Code Copilot, Cursor — connect to the server and call its tools during a conversation. The AI decides when to call find_callers before proposing a rename, or get_diagnostics after generating code. You don’t have to prompt it; the tool availability shifts how it reasons.

Getting Started

Install as a .NET global tool:

dotnet tool install -g RoslynCodeLens.Mcp

Or install as a Claude Code plugin:

claude install gh:MarcelRoozekrans/roslyn-codelens-mcp

The server is also listed on Glama.ai and available on NuGet.

Source and full tool reference: github.com/MarcelRoozekrans/roslyn-codelens-mcp