io.github.DiegoBr4nd/godot-gut-mcp
<!-- mcp-name: io.github.DiegoBr4nd/godot-gut-mcp -->
# Godot GUT MCP
Servidor MCP que permite a asistentes de IA (Claude, Cursor, etc.) ejecutar las
pruebas unitarias de tu juego en Godot usando [GUT](https://github.com/bitwes/Gut),
y leer los resultados de forma estructurada. Ideal para el ciclo
"la IA escribe código → ejecuta los tests → lee los fallos → corrige".
## Requisitos
- Python 3.10 o superior
- Godot 4.x
- [GUT 9.x](https://github.com/bitwes/Gut) instalado en tu proyecto (carpeta `res://addons/gut/`)
## Instalación
```bash
pip install godot-gut-mcp
```
## Configuración en Claude Desktop
## Configuración en Claude Desktop
Añade esto a tu archivo `claude_desktop_config.json`:
```json
{
"mcpServers": {
"godot-gut": {
"command": "uv",
"args": ["run", "--with", "godot-gut-mcp", "godot-gut-mcp"],
"env": {
"GODOT_PATH": "C:/ruta/a/godot.exe",
"GODOT_PROJECT_PATH": "C:/ruta/a/tu/proyecto"
}
}
}
}
```
- `GODOT_PATH`: ruta al ejecutable de Godot.
- `GODOT_PROJECT_PATH`: ruta a la carpeta de tu proyecto (la que contiene `project.godot`).
## Herramientas disponibles
| Herramienta | Qué hace |
|---|---|
| `run_all_tests` | Ejecuta toda la suite de pruebas y devuelve un resumen |
| `run_test_file` | Ejecuta un archivo de pruebas concreto |
| `get_failures` | Devuelve solo las pruebas que fallaron, con el mensaje y la línea del error |
## Ejemplo de respuesta
```json
{
"summary": { "passing": 1, "failing": 1, "all_passed": false },
"tests": [
{
"test": "test_suma_basica",
"status": "failed",
"message": "expected to equal [2]: Dos más dos debe ser cuatro",
"line": 4
}
]
}
```
## Licencia
MITTDQS
Scored across 3 tools
Each tool has a clear, distinct purpose: running all tests, running a specific test file, and retrieving failures from the last run. There is minimal overlap, as get_failures is a targeted diagnostic rather than a duplicate of the test runners.
All tool names follow a consistent verb_noun snake_case pattern: run_all_tests, run_test_file, get_failures. No mixed conventions or vague verbs.
With 3 tools, the server is slightly lean but well-scoped for its purpose of running GUT tests and reporting failures. It sits at the lower edge of the ideal range but does not feel insufficient.
The server covers the core testing lifecycle: run all tests, run a specific file, and inspect failures. A minor gap is the lack of running a single test method or listing available tests, but this does not severely hinder typical usage.