A coding agent can write C# without much trouble.
The expensive part starts when it needs to understand your Unity project, control the Editor, enter Play Mode, inspect the result, and fix what it broke.
MCP made that possible, but large MCP toolsets can consume tens of thousands of context tokens before the agent has done any useful work. Every exposed tool adds instructions, schemas, and more opportunities for the agent to pick the wrong thing.
Unity CLI changes the shape of that workflow.
Instead of loading every Unity tool into every turn, the agent gets one command-line frontend and a small set of project-specific skills. It loads the detailed instructions only when the task needs them.
The result is less context overhead, fewer irrelevant tools, and one interface for Editor control, scenes, assets, screenshots, runtime evaluation, builds, and automation.
This module shows the setup and puts it to work with Pi: a lean coding-agent harness that uses Unity CLI to create a gameplay feature, place it in the scene, test it in Play Mode, inspect the result, and catch a real bug.