Making games from Cline

People who choose Cline usually chose it on purpose: it is open source, it runs in an editor they already have, and they supply their own model key so the model cost is theirs to see and control. That is a coherent position and we are not going to argue you out of it.

It does mean you should know what connecting to Stage Engine costs on our side. External MCP access is part of the paid plans. Free is $0, Indie is $10 a month, Pro is $50. So the arrangement is your model bill plus a Stage Engine subscription, and anyone telling you it is free is not counting properly.

What the paid tier buys is the outside access and more usage. It is not a different engine, a different builder, or a better version of the capability layer. The layer is the same one Director 1 uses on the free plan.

There are two MCP servers inside the Stage Engine desktop app: one exposing the Stage capability layer, one exposing assets. Cline is an MCP client. Both sides implement the same protocol, so they can talk. That is the entire integration story and we would rather state it flatly than dress it up.

We do not publish a Cline extension, we do not maintain a listing, and there is no partnership. Any client that speaks MCP should work. These are the popular ones people ask about, we have not tested each of them, and behaviour will vary.

For the actual connection steps, use the app and its documentation. Client configuration moves between versions and a stale example on a landing page costs you an evening.

Agents that separate planning from execution are only as good as the vocabulary they plan in. If the available operations are things like set a property or add a node, a plan reads as a list of keystrokes and tells you nothing about whether the result will be the game you wanted.

The capability layer gives a coarser and more meaningful vocabulary. Terrain of a given shape. Water following an outline you drew. Weather, sky, characters and their behaviour. A plan written in those terms is a plan you can read and disagree with before anything happens, which is the point of having one.

It also cuts the amount the model must invent. Every capability is code that already exists and already works, so composing against it is a smaller job than writing it from nothing. That is why the layer matters more than which model you pointed at it.

You build the world by hand, in the builder inside the app. Placing terrain, water, weather, buildings and characters is work done with your hands and your eyes, because that is where taste lives and because describing a hillside is slower than shaping one. No agent takes that over.

You also decide whether the game is right. You play it, you dislike something, you go back in. That loop is the normal way to work here. Not writing the code has never implied not judging the result, and nothing about bringing your own agent changes who the judge is.

If shuttling between an editor and the app sounds like friction you did not want, using Director 1 inside the app is a completely reasonable answer, and it needs no setup at all.

Stage Engine is a desktop application for Mac and Windows. Downloaded, installed, with its own copy of Godot inside it. No browser version, no hosted editor. The project it works on is a real Godot project in a folder on your machine, so your agent operates on files that exist and version control behaves normally.

When you press play, the game runs locally on the bundled Godot at your machine's framerate. Nothing is streamed and nothing is rendered somewhere else.

Building needs an internet connection. The AI, asset generation and world data all come over the network, and that is true no matter whose agent is driving. Playing a game you already built does not need it.