Google Playground turns prompts into playable games, but it only works in the US. Here is what it does, and the agent toolkit for art, 3D, music and SFX that works anywhere.
01What Google Playground is
On October 7, 2026 Google Labs launched Playground, an experimental platform where you create games by chatting. You describe the idea, pick a starting genre (trivia, tower defense, racing, platformer, arcade shooter and others) or start from a blank canvas, and Playground builds a playable 2D or 3D game in your browser. Games can be single-player or multiplayer, have leaderboards, and can be kept private, shared by link or listed in a community gallery.
According to the help center, you can also upload your own PNG or JPG assets and export the finished game as a ZIP with its HTML, JavaScript, CSS and generated assets. Google does not name the models in its post; 9to5Google reports that it uses Gemini, Nano Banana and Lyria. A Unity integration for higher-fidelity 3D is announced for later in 2026.
02Why you can’t use it in Brazil (yet)
Playground launched for users in the United States who are 18 or older. Everywhere else, including Brazil and the UK, the site redirects to a "Not available in your region" page that says it is "expanding to more regions soon". Google has not given a date. Inside the US, playing is free and creating uses a free tier of weekly tokens, with higher limits on a Google AI plan.
If you are outside the US and want the same "describe it, play it" experience from Google, two official options work in Brazil today: Gemini Canvas, which builds small apps and games inside the Gemini app (in Portuguese, free), and Google AI Studio Build mode, which generates a full web app you can export or deploy.
03The other route: build the game yourself, with an agent
Prompt-to-game platforms are great for prototypes, but the game lives on their platform and inside their limits. The route most indie developers are taking in 2026 is different: you build the game in your own engine or code (Godot, Unity, Phaser, plain web) with a coding agent such as Claude Code or OpenAI Codex, and you connect AI tools to that agent through MCP, the Model Context Protocol.
With MCP, the agent can call each tool directly: "draw the player sprite", "make a low-poly treasure chest", "write the boss theme", "add a coin pickup sound". The files land in your project folder, you own the code, and nothing depends on a region launch. These four tools cover art, 3D and audio, and all of them have an MCP route:
- PixelLab
pixel-art characters, animations, tilesets and UI — official hosted MCP - Tripo
text or image to 3D, with an official CLI skill for Claude Code and Codex - Meshy
text or image to 3D, retexturing, rigging and animations — official MCP - SunoBoard
music, loops and sound effects (Suno, Lyria, ElevenLabs and more) — hosted MCP
04Pixel art: PixelLab
PixelLab generates pixel-art characters in 4 or 8 directions, animates them, and builds top-down and side-scroller tilesets, map objects and UI. It has an official hosted MCP server, so Claude Code can create a character and its walk cycle while it writes the movement code. There is a free trial (40 fast generations, then 5 a day), and at the time of writing PixelLab also offers a lower-priced entry plan in Brazil. Full setup in our PixelLab guide.
claude mcp add pixellab https://api.pixellab.ai/mcp -t http -H "Authorization: Bearer YOUR_PIXELLAB_KEY"
053D models: Tripo and Meshy
Tripo and Meshy both turn a prompt or an image into a textured 3D model you can export as GLB or FBX for Godot, Unity or Unreal, and both can rig and animate characters. Meshy ships an official MCP server you run with npx (it needs a Pro API key); Tripo’s current agent route is an official CLI that installs a skill for Claude Code and Codex. Watch the licences: free-tier output is non-commercial on Tripo and CC BY 4.0 on Meshy. Details in our Meshy and Tripo guides.
06Music and sound effects: SunoBoard
Art and models are only half a game. SunoBoard gives your agent music and sound effects through one hosted MCP server: title themes, level loops and boss fights with Suno, Google Lyria, MiniMax or ElevenLabs Music, plus footsteps, coins, doors and UI clicks in about two seconds with ElevenLabs SFX. In Claude.ai you add it as a custom connector with no API key; in Claude Code you add the URL with your SunoBoard key.
In OpenAI Codex CLI, add it as a remote MCP server in ~/.codex/config.toml (Codex supports URL servers with a bearer token read from an environment variable). Prefer a local server? The open-source package @yellowkode/suno-mcp does the same over stdio. See real released games scored this way.
claude mcp add --transport http sunoboard https://mcp.sunoboard.com/mcp --header "Authorization: Bearer sb_YOUR_KEY"
# ~/.codex/config.toml [mcp_servers.sunoboard] url = "https://mcp.sunoboard.com/mcp" bearer_token_env_var = "SUNOBOARD_KEY"
07One prompt to wire the whole pipeline
Once the servers are connected, a single request can use all of them. Ask for small batches and review as you go: the agent works fastest when each asset has a clear job in the game.
We are making a cozy 2D farming game in Godot. Use PixelLab to create the farmer character (4 directions, walk animation) and a top-down grass and dirt tileset. Use SunoBoard to generate a calm acoustic day theme that loops, a night theme, and sound effects for footsteps on grass, watering a plant and a coin pickup. Save everything under assets/ and wire the sounds to the player script.
Use Meshy to make a low-poly treasure chest (GLB, under 10k triangles) with an open animation, and SunoBoard to make a short wooden chest-opening sound effect and a 5-second victory jingle. Import them into the Unity project and trigger both when the chest opens.
