See a real worked example of building a stateful multiplayer app on the mcp-gtw gateway library
Test how an AI agent controls a character using a small, well-defined set of MCP tools
Study how a single process can serve an MCP endpoint, game simulation, and web client together
Use as a starting point for building your own AI-controlled real-time game or simulation
| mcp-gtw/demo-game | 0xustaz/streamgate | a-bissell/unleash-lite | |
|---|---|---|---|
| Stars | 1 | 1 | 1 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | hard | hard |
| Complexity | 3/5 | 4/5 | 4/5 |
| Audience | developer | developer | researcher |
Figures from each repo's GitHub metadata at analysis time.
Requires Python 3.12+, the uv tool, and Node 22+ with npm to build the Vite client bundle.
MCP Game is a demonstration project showing off what is possible with a library called mcp-gtw. It is a multiplayer, top down action game where every character in the game is controlled entirely by an AI agent, not by a human clicking or typing directly. There are no manual keyboard or mouse controls at all: a person opens the web page in a browser, that page hands their AI agent a private connection, and once the agent logs in, it becomes that character in the shared game world. The AI controls its character using a small set of tools such as move, look, attack, shoot, chop, and speak, which are the only ways to affect the world. The server keeps full authority over the game, calculating everything in terms of a grid, so all agents are playing fair by the same rules rather than trusting anything the browser or the agent claims. When a player logs in, they can choose a character class such as warrior, archer, monk, or lancer, along with a skin color, while enemy characters always appear in red. Under the hood, the whole thing runs as a single program that serves the connection for AI agents, the authoritative simulation of the game world, a live view stream, and the game's own web based client all at once. The visual side uses the Phaser game framework along with a set of licensed 2D art assets, and the project bundles everything so it can be run locally without hunting down separate pieces. To try it, you need Python 3.12 or newer, the uv Python tool, and Node.js with npm to build the web client, after which a couple of setup commands install everything and start a local server in a browser. This project is aimed at developers exploring how AI agents can interact with real time, stateful applications through a standard protocol, and it works well as a hands on example of the mcp-gtw library rather than a game meant for casual players.
A multiplayer action game where every character is played entirely by an AI agent through MCP tools, demonstrating the mcp-gtw library.
Mainly Python. The stack also includes Python, Phaser, Vite.
MIT license: free to use, modify, and distribute for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.