// the take
An MCP server that bridges AI coding assistants to the Unity Editor, exposing scene authoring, builds, profiling, and multiplayer (MPPM/ParrelSync) controls as callable tools. It's aimed at Unity developers who want Claude or Cursor to directly manipulate a running editor rather than just generate code, and specifically targets setups running multiple editors or agents at once.
Multi-editor support is actually architected, not bolted on: per-editor request queues, port-based discovery that verifies live Unity identity before binding, and explicit project/agent selection so simultaneous agents don't clobber each other's routing. The tool surface is tiered (a small set of ~80 exposed tools plus ~269 discovered on demand via search+schema+invoke), which avoids blowing out a client's tool-count limit while still exposing the full API. Test coverage is real: 263 server tests run across 8 CI configs (Node 18-24 on Windows/Linux) against mock bridges, backed by linked real-editor validation logs rather than bare claims. MPPM and ParrelSync support covers a genuinely underserved case — most Unity AI-bridge projects ignore multiplayer play-mode testing entirely.
Ships under a custom 'AnkleBreaker Open License v1.0' rather than a standard OSS license, with explicit restrictions on reselling — worth reading closely before using this in anything commercial, since it isn't MIT/Apache-style permissive. The README documents the unreleased 'modernization branch' and tells you to clone main directly because 'published npm packages and version tags may lag behind main,' so the stable release path is behind what's documented and you're expected to track main yourself. Real-editor validation is Windows-only on a single Unity version (6000.6.2f1); the claimed 2021.3.18f1+ compatibility and non-Windows/IL2CPP builds are explicitly untested, so the compatibility story is mostly aspirational outside that one configuration. It also openly states that concurrent agents can make conflicting edits to the same scene object with no resolution mechanism — you have to manually partition ownership, which is a real operational burden once you're actually running the multi-agent setup this project is built around.
We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →