Making games from Gemini CLI

A capable agent pointed at a Stage Engine project can read all of it. The project is a real Godot project in a folder on your machine, in ordinary text formats, so there is nothing to unpack and nothing hidden. Ask what a script does and you will get a decent answer.

That feels like most of the problem solved and it is not. Understanding a project tells you what is there. It does not tell you what should be there, and it does not give you a safe way to put it there. An agent that has read every file still has to decide how to express a change, and in a bare game project the only vocabulary available is fine-grained: create a node, set a property, write a script.

Those are keystrokes, not operations. A model composing at that grain can be wrong in a hundred small ways that each look reasonable when you inspect them one at a time.

Stage Engine ships two MCP servers inside the desktop app: one exposing the Stage capability layer, one exposing assets. The layer is a set of operations at the level a game is actually described in. Terrain of a given shape. Water following an outline. Weather, sky, characters and what they do.

Each capability is a body of code that already exists and already works, which means it is a body of code the model does not have to invent. That is the whole bet. Less invention means fewer failures, and it means a cheaper model composes reliably where a bare project would have required an expensive one just to keep the plausible mistakes down.

So the interesting difference is not context size or how much of your project an agent can hold at once. It is whether the surface it is calling was designed for games or is just the engine's internals with a protocol taped on.

Worth being direct, because the low entry cost is often exactly why people picked this tool. External MCP access to Stage Engine is part of the paid plans. Free is $0, Indie is $10 a month, Pro is $50. Whatever your agent costs you is separate and unrelated.

The Stage Engine download is free and there is a free plan, and on it you use Director 1, which ships inside the app and needs no setup. That is the free path. Driving the app from outside is the paid one.

What the paid tiers buy is usage and that external access. Not a better engine, not a better builder, not a different capability layer.

We have not built anything specific to Gemini CLI. We speak MCP; it speaks MCP. There is no extension we publish, no listing, no partnership. Any client that speaks the protocol should work, and the ones people ask about are simply the popular ones. We have not tested them all and behaviour varies.

For the connection itself, follow the app and its documentation rather than a snippet on a marketing page. Client configuration changes between versions and a wrong example is worse than none.

You build the world by hand in the builder inside the app: terrain, water, weather, buildings, characters, placed and looked at. That is where taste is the input, and it does not move to an agent because an agent is connected. It is the design, not a gap waiting to be filled.

You also judge. You play the game, decide what is wrong, and go back in. That loop is the normal work here rather than a sign the tooling failed. Not writing the code has never meant not having an opinion about the game.

Stage Engine itself is a desktop application for Mac and Windows with Godot inside it. Games run locally when you press play. Building needs an internet connection because the AI, asset generation and world data come over the network; playing what you already built does not.