AI MCP Bridge

AI MCP Bridge turns any MCP-compatible AI assistant like Claude Code, Claude Desktop, Cursor, and others into a hands-on collaborator inside your Unity Editor. Instead of describing changes, your agent makes them: it creates and edits GameObjects, wires components, builds prefabs and variants, sets up animators, particles, UI, lighting and post-processing, manages render pipelines, runs builds and tests, and profiles performance all against your live, open project. It's not tied to any genre or art style and ships no content to customize. It's a workflow and automation layer that works on whatever project you already have 2D, 3D, mobile, VR/XR, URP, HDRP or Built-in.

How it works: the Unity asset hosts a local WebSocket server (127.0.0.1) that your AI client connects to through a small Node-based MCP server (shipped in-package as plain JavaScript, run by your own Node.js). Every request is marshaled onto Unity's main thread, so there are no threading surprises, and the bridge auto-reconnects across domain reloads and editor restarts. Multiple clients can share one editor. All render-pipeline access is reflection-based, so the package never hard-depends on URP/HDRP/XR packages — features light up when the relevant package is installed and degrade with a clear message when it isn't.

An optional built-in sub-agent (unity_agent) can run an autonomous tool-calling loop using your own OpenAI, Anthropic, Gemini, OpenRouter or Flowdot key (no keys bundled or required). API keys are read from your MCP client config and redacted from all logs.  This is so that you can manage context - you can ask your main agent to call the sub-agent for specific tasks, or let the sub-agent run autonomously doing some context heavy task.

A built-in unity_learn knowledge base documents every subsystem render pipelines, 9-slice UI, profiling playbooks, lighting, and a running "gotchas" list so the agent works correctly the first time.