MCP Tooling
The Model Context Protocol is a shared wire format for giving AI assistants access to external tools. Dynamo uses it to drive your game engine: install the engine package and the editor's tools show up in chat, prefixed mcp__<engine>__<tool> (e.g. mcp__unity__load_scene).
Today the supported integration is Unity, shipped in the @dynamoengine/unity package.
Driving the Unity Editor
The companion `@dynamoengine/unity` package gives the assistant direct control of a running Unity Editor — and the connection stays live in Play Mode, so the model can drive and inspect your game while it runs. Roughly 76 tools (scenes, assets, play-mode input, runtime assertions, tests, builds, settings) become available in chat once the bridge is connected. See Unity Integration for the full setup.
A tool's permission level is derived from the annotations it declares: read-only tools auto-execute, destructive tools land in the Dangerous tier, and the rest prompt once per tool, per model. Dangerous tools prompt and never offer "Allow all" — but Auto Mode approves the whole tier, so leaving a long task running unattended means the bridge's destructive tools run without asking. See Permissions for what Auto Mode does and doesn't gate.
Connecting & managing
- Startup — Dynamo reports the bridge state on launch: connected (with a tool count), reachable but the Unity Editor isn't responding, or not installed (with an install hint).
- `/mcp` — live bridge state, transport, tool/resource/prompt counts, and backend readiness; press
rto reconnect. - `/connect` — reconnect the engine bridge (handy after you start or restart the Unity Editor).
The model can manage the connection too, via admin tools: mcp_list_servers, mcp_server_reconnect, mcp_tool_enable / mcp_tool_disable (session-scoped), and mcp_resource_read / mcp_prompt_get for fetching the bridge's resources and prompts.