I made an open-source MCP server for Godot that gives AI agents free vision, image generation, and the ability to actually play and test your game.
The biggest thing: built-in Agnes + NVIDIA AI APIs — all free. The AI can see your game screenshots and understand them, generate sprites/textures/art for
your project, even generate videos. No other Godot MCP has this. No API costs — these are free endpoints with multi-key rotation built in.
That means a complete loop nobody else offers: AI plays the game → takes a screenshot → actually sees and understands the frame → figures out what’s wrong ->fixes it → generates new art if needed → re-tests.
Also unique:
• Deterministic playtesting — freeze the clock, step time, inject inputs. AI reproduces bugs and verifies fixes instead of guessing.
• Multiplayer testing — host + client instances, latency injection, sync validation. Only Godot MCP with multiplayer support.
• DAP debugging — real breakpoints, step over/into, variable inspection.
• LSP integration — static diagnostics, autocomplete, go-to-definition.
• Token-efficient — JSON state digests instead of screenshots (saves ~90% tokens), diff-only returns.
Plus the usual: scene/node/script editing, profiler, C# support, smart reconnect, in-editor dock, Windows orphan process cleanup.
100% MIT, no paywall.
Links:
• GitHub: GitHub - masteryee-labs/Open-Godot-MCP: Open-source MCP server for AI-autonomous Godot game development. Deterministic playtesting, multiplayer testing, DAP debugging, LSP, token-efficient. 100% MIT. · GitHub
• Asset Library: Open Godot MCP - Godot Asset Library
• pip install open-godot-mcp
v0.1.8 now. Try it and let me know your honest opinion — what works, what breaks, what’s missing!