Summer Engine MCP guide
How to make games with Ollama and Summer Engine
Run a free local model with Ollama and drive Summer Engine through MCP — a real desktop game engine, no AI subscription required. Summer Engine is the desktop game engine; Ollama drives it through MCP with a model running on your own hardware.
Summer is free for MCP use
Download the desktop engine and connect it to your agent. The model runs on your own hardware — no AI subscription, no per-message cost.
Open Ollama
Run a free local model on your own GPU. The setup writes the MCP config so {agentLabel} can drive the engine directly.
Launch a real game
The local model writes code. Summer Engine shows the game, runs Play mode, and exports projects for Steam, desktop, mobile, and console pipelines.
Local setup notes
The 2026 specifics.
Pair Ollama with an MCP agent
Ollama serves the model but is not an MCP client by itself. Pair it with OpenCode (terminal), Kilo Code (VS Code), or Goose — the setup command above wires OpenCode, the most common pairing in 2026. Point its provider at http://localhost:11434/v1.
Fix the context window first
Ollama defaults to a 4k context on GPUs under 24 GB of VRAM, and MCP tool schemas alone can overflow that — the failure is silent. Start the server with OLLAMA_CONTEXT_LENGTH=65536 (or raise the slider in the Ollama app), then run `ollama ps` to confirm the model still fits on the GPU.
Models that hold up
Tool calling is the hard part for local models. Safe picks in mid-2026: gpt-oss-20b on 12–16 GB cards, Qwen3-Coder-30B or Gemma 4 27B on 24 GB. Use Q4_K_M quantization or better — heavier quantization breaks tool calls before it breaks chat.
Why this matters
Ollama runs free on your hardware. Summer Engine shows the game.
A capable local model can operate a real engine through MCP: open scenes, place objects, run the project, and read errors. Your GPU does the thinking — no subscription, no per-token bill, and your prompts never leave your machine.