Unity Control Protocol

Modern agentic development assumes the tool you are working in has a structured, programmatic interface. The Unity Editor was never designed for that. If you wanted an agent to drive a Unity project, your options were: scripting through the Editor's GUI automation surface (slow, brittle), or pretending the Unity project was just a folder of YAML files (loses too much).

UCP fixes this. It turns the Editor into a JSON-RPC server with first-class structured access to scenes, GameObjects, components, materials, prefabs, build pipelines, tests, packages and the profiler. From the outside, the Editor stops being a black box.

The shape is intentionally simple. A Rust CLI (the user-facing tool) and a C# bridge inside the Editor talk over a WebSocket with JSON-RPC 2.0. Every Editor action that has a programmatic counterpart gets exposed as a method. Every method has a well-typed input and output. The protocol is small enough to audit.

In practice, that means an agent (or a human at the terminal) can:

• Inspect, create and modify scenes and GameObjects.
• Add and configure components, materials, prefabs.
• Selectively import .unitypackage contents instead of dumping everything.
• Run builds, tests, asset processors, packages.
• Set Editor settings, hook into debugging and profiling.

UCP ships as a Rust crate and a Unity package, plus - usefully - a Claude Code skill, so any session inside Claude Code that needs to touch a Unity project can drop into the protocol without any extra setup. That last detail matters: the friction of getting from "agent" to "Editor" was the whole reason I built this in the first place.