We open sourced our Unity MCP server

Many “AI for game dev” tools still stop at code generation.

They can suggest a script, maybe explain an error, maybe even produce something close to what you want. But in actual Unity workflows, that is usually only a small part of the job.

The real work is spread across scene hierarchy, prefabs, materials, UI, physics, animation, input setup, package differences, console errors, project conventions, and lots of repetitive editor actions.

That gap is exactly why we built GladeKit.

Today, we’re doing two things:

1. Launching GladeKit officially (see Product Hunt)


2. Open sourcing the GladeKit Unity MCP server
   

GladeKit Unity MCP

The open-source MCP server connects AI clients like Cursor, Claude Code, and Windsurf directly to the Unity Editor.

That means the model is not just chatting about your game in the abstract. It can actually operate with real Unity context.

The server includes:

• 230+ Unity tools across areas like scenes, GameObjects, scripts, prefabs, materials, lighting, VFX, audio, animation, physics, camera, UI, input, terrain, and NavMesh


• a Unity-aware system prompt


• GLADE.md project context injection


• semantic script search


• skill calibration based on user expertise


• optional cloud intelligence for RAG and cross-session memory

Core features are free, local, and MIT licensed.

Why we open sourced it

For Unity especially, usefulness depends on project awareness. The model needs to understand what scene is open, what objects exist, what scripts are relevant, what pipeline is being used, what errors are happening, and what conventions the project already follows.

Without that, you end up with generic “AI-generated advice.”

With that, you start getting closer to an actual useful AI assistant / agent.

Open sourcing the MCP server is our way of pushing that interface forward.

Example of the difference

A normal coding assistant might help with:

“Write me a script for enemy spawning.”

A Unity-connected MCP can help more like this:

“Find how enemy spawning currently works in my project, inspect the related scripts, create a new spawn manager, wire it into the scene, and adjust the exposed values to match the existing design.”

That difference is what we care about.

Architecture at a high level

The setup is simple:

• a Unity bridge package runs inside the editor


• the MCP server connects to that bridge


• your AI client talks to the MCP server over stdio or HTTP


• the model gets tool access plus Unity-specific context

So instead of copy-pasting back and forth between your IDE, a chatbot, and Unity, the agent can operate much closer to the actual source of truth.

Why this matters beyond GladeKit

I think game dev is one of the most interesting places for MCP-style tooling.

Game development has a huge amount of structured-but-fragmented work:

editor actions, asset references, scene state, component wiring, engine-specific APIs, and long chains of small tasks that are annoying to do manually but difficult to solve with plain text generation alone.

That makes it a really good fit for agent tooling with real tool access.

My guess is we’ll see more of this pattern across game engines and other developer tools - not just AI that answers questions, but AI that can actually operate in the environment where the work is happening.

Links

Open-source MCP repo:

https://github.com/Glade-tool/glade-mcp-unity

GladeKit site:

https://gladekit.com

Product Hunt launch:

https://www.producthunt.com/products/gladekit?launch=gladekit

Would love feedback from anyone building AI dev tools, working with MCP, or trying to make Unity workflows faster.