Turned VS Code / Cursor into a AAA Unreal Engine C++ IDE (Live Execution, Blueprints as C++, AI MCP Server)

Hey everyone! :waving_hand:

Like many of you, I love Unreal but was getting frustrated with how heavy Visual Studio can be, and I didn’t want to switch my entire workflow to Rider. So, I spent the last few months building UStudio — an extension that gives VS Code and Cursor true engine awareness.

I wanted to share it with the community because it solves some of the most annoying C++ pain points:

:high_voltage: Live Execution: You can actually watch your C++ code light up and count calls in real-time while playing in the editor, exactly like the Blueprint Debugger. :puzzle_piece: Read Blueprints in VS Code: It parses binary .uasset files directly. You can open any Blueprint as a C++ header to see its components, variables, and how they use your code, without launching the engine. :robot: Local AI MCP Server: If you use Claude or Cursor, you know AI hallucinates with UE macros. I built a local Model Context Protocol server that feeds your AI the real context of your Blueprints, reflections, and Build.cs. :hammer_and_wrench: Smart Fixes: It diagnoses UHT and linker errors, and gives you an interactive multi-file diff preview before applying #include or module fixes.

:television: I made a quick 90-second tour showing it in action here: https://www.youtube.com/watch?v=_aXXyKU8Oto

It comes with a 14-day fully featured free Pro trial. I’d love for you hardcore C++ devs to test it out and tear it apart. Feedback is super welcome!

:link: Download (Open VSX / VS Code): Open VSX Registry :speech_balloon: Discord: G-Unreal Engine Tools

Nice work, reading Blueprint assets from VS Code without launching the editor is a real time saver. A few things worth documenting for people who try it:

  1. Which engine versions the .uasset parser has been tested against. The package format changes between releases, so a short compatibility table (5.3, 5.4, 5.5 and so on) would save a lot of support questions.
  2. How Live Execution counts calls: through lightweight instrumentation in the editor session or by hooking at runtime. Knowing the overhead matters for anyone profiling a busy gameplay loop.
  3. For the local MCP server, how much context it sends per request. Large projects can have thousands of reflected classes, so filtering by module or by the files currently open will keep AI suggestions accurate and fast.

A small note on Build.cs parsing would also help, since modules with conditional dependencies per platform tend to confuse most tooling. Thanks for sharing it with the community.

1 Like

Thank you so much for the feedback! You bring up exactly the kind of technical questions that matter for production environments. I’ll definitely make sure to highlight these details more prominently in the documentation.

To answer your points:

1. Supported Engine Versions: The .uasset binary parser (along with the rest of the extension) is fully tested and supported across Unreal Engine versions 5.0 through 5.8. (Just a small note: the one-click Blueprint variable generation via Python requires 5.6 or newer, but parsing works across the board).

2. Live Execution Overhead: It uses explicit, lightweight instrumentation rather than heavy runtime hooking. When you choose to trace a function, UStudio inserts a USTUDIO_TRACE() macro at the top of it. This costs only a few atomic operations per call and completely compiles out to nothing in Test and Shipping builds, so it won’t pollute your profiling data in release candidates.

3. MCP Server Context Limits: The server doesn’t dump the entire codebase into the AI’s context window. Instead, it exposes targeted tools to the agent (like read_blueprint_as_cpp, find_references, or find_header_for_type). The AI actively queries only the specific classes or files it needs at that moment, keeping the token usage low and the suggestions highly accurate without overwhelming the LLM.

4. Build.cs & Conditional Dependencies: Good call on this. The extension parses Build.cs intelligently. For example, when its quick-fix system detects a missing editor-only module, it automatically wraps the dependency injection inside an if (Target.bBuildEditor) block rather than blindly appending it to the public dependencies. It also leverages UBT sources to provide proper autocompletion for TargetRules and ModuleRules.

Thanks again for taking the time to write this up! Let me know if you run into any other edge cases.