MCP Server for Phaser Editor v5

This Model Context Protocol (MCP) server provides the tools to connect an LLM with Phaser Editor v5. It is designed to work seamlessly with Phaser Editor’s IDE features, allowing developers to create and manage game scenes efficiently.

Installation

This server is available as an NPM package and you can use it in the configuration of MCP hosts like Cursor, Claude Desktop, GitHub Copilot, etc…

This is an example of a configuration file for Cursor:

{
    "mcpServers": {
        "phaser-editor": {
            "command": "npx",
            "args": ["@phaserjs/editor-mcp-server"]
        }
    }
}

This servers connects with a running instance of the Phaser Editor v5 desktop application. All changes on the project are made through the Phaser Editor running instance.

Coding

All tools provided by this server are specifically tailored to work with Phaser Editor features, such as the Scene Editor, Asset Pack Editor, and others. These tools are not intended for direct modification of your project’s source code. However, when used alongside code editors like Cursor, the server enables simultaneous interaction with both Phaser Editor’s custom files and the project’s source code—creating a powerful, integrated development workflow.

We recommend starting with one of the Phaser Editor project templates, which are preconfigured for seamless integration with Cursor. These templates include all necessary configurations and the rules for the LLMs to understand how to interact with Phaser Editor’s custom files and the project’s source code.

Using Training Levels to Enhance LLM-Based Content Generation

Large Language Models (LLMs) excel at generating new content when provided with high-quality, contextual examples. This is especially effective for game level generation. To take full advantage of this capability, it’s a good practice to create training levels that clearly express the concepts and elements you want the model to replicate.

Why Training Levels Matter

Training levels serve as concrete examples that define the structure and logic of your game. These examples help the LLM understand the components involved and how they interact, enabling it to generate new levels that align with your design.

For example, if you’re working with tilemap-based levels, you can create a TrainingLevel scene that includes the key objects used in your game. To improve clarity and semantic understanding:

• Use one layer per object.
• Name each layer with a descriptive identifier, such as pipe, cloud, or mountain.

This organization helps the LLM detect patterns and apply them to new content generation tasks.

Visual Tools Help—but Context is Key