Unity's Official Claude Code Plugin: 29 Skills, Plus Extra Setup for Live Editor Control

The Unity plugin for Claude Code is a first-party set of 29 Unity skills that teach Claude Code how Unity's own teams want common jobs done, from RuleTiles to UI Toolkit to Unity IAP. It installs with two slash commands, but it ships skills only. Live control of an open Editor needs the experimental Unity CLI, the experimental Pipeline package in a Unity 6+ project, a signed-in account, and the Editor running.

That last point is where the launch coverage and Unity's documentation disagree. This Lab Note sets out what the plugin contains, what it leaves out, and the full checklist for the Editor-control part.

What did Unity ship, and when?

Unity announced the plugin on Sep 9, 2026, in a blog post by Rachel Zhao, Director of AI Products. It installs from the Claude Code terminal or from Claude Code inside the Claude Desktop app. The Claude plugin directory listing names Unity Technologies as the maker and showed 485 installs at the time of writing. That figure is a snapshot, not a trend.

Key facts from the sources:

• 29 skills at launch, with more promised, according to the blog and the Unity Discussions announcement.
• Other agents are already listed. The blog said other agents would follow "later this year", but the GitHub README and Unity's About page already cover Codex and Grok. Unity staff confirmed the Codex version in the forum thread.
• Beta version, source-available license. The README's sample codex plugin list output shows version 0.1.0-beta, and the repo uses the Unity Companion License.
• Unity 6.0 or later only, per both the docs and the README.
• Not a Unity package. The About page says it is not an Asset Store item and does not install through the Package Manager.

For more coverage of the engine and agent tooling, see /tags/unity and /tags/coding-agents.

What does the plugin install, and what doesn't it?

The blog's meta description says the plugin installs skills, the Unity CLI, and live Editor control "via Unity's MCP server, in one command." Unity's own Claude Code docs page says otherwise. Its troubleshooting section states the plugin "only ships skills and doesn't use" an MCP server. The README agrees: skills only, "no hooks and no MCP servers."

How the pieces actually fit, according to the docs:

• Skills are packaged instructions plus reference material that load when a task matches. Each skill keeps detailed guidance in a references/ folder, per the About page.
• Editor control happens when skills run Unity CLI commands. The /unity-cli skill teaches Claude how to drive the CLI. It does not install the CLI binary.
• Unity's MCP server is a separate product, set up with unity mcp configure.

Unity argues that skills lead to fewer errors, better token efficiency and fewer turns than letting Claude improvise. The blog offers no published benchmark for that, so treat it as Unity's claim.

The 29 skills, by area, and where do they lean?

Counts below come from the skill table in Unity's announcement.

Details from the table worth knowing:

• /new-unity-project runs a guided setup but "does not scaffold gameplay code."
• /ui detects which UI system a project already uses before writing any UI code.
• /audio-setup-mixers reports Audio Sources it couldn't match instead of skipping them silently.
• /ui-imgui is labeled maintenance only. New editor UI defaults to UI Toolkit.

The 2D tileset pairing

For 2D developers, the clearest fit is two skills built to work together. According to the blog, /sprite-segment-3x3grid splits a sprite into a 3×3 grid by colour match and outputs a text pattern. /tilemap-ruletile-createfromsegment reads that pattern and builds auto-tiling RuleTiles, sorting neighbour rules by specificity. Wiring RuleTile neighbour rules by hand is slow, so on paper this is the most concrete time-saver in the set. More tile workflows live under /tags/tilesets.

Where the coverage leans toward Unity's own stack

Five skills target Unity's own services and SDKs: Unity IAP, LevelPlay, Vivox, /build-live-game (on Unity Services) and Multiplayer Services. The IAP skill also covers migrating off RevenueCat and Adapty, among others. These skills may well be useful, but part of this coverage clearly exists to move projects onto Unity's commercial stack.

Gaps

The launch table lists no skills for animation, input, general gameplay scripting patterns, or HDRP. URP is the only render pipeline with rendering skills.

How do you get live Editor control working?

"One command" covers step 1 only. Steps 2 to 4 are separate installs, and Unity labels the components they add as experimental.

1. Install the plugin. Inside a Claude Code session, run:

   /plugin marketplace add Unity-Technologies/unity-agent-plugin
   /plugin install unity@unity-agent-plugin
   

   The docs note that /plugin isn't available in the VS Code extension. Use /plugins there to open the plugin manager. Choose a scope: user (all projects), project (shared through the repo, suited to teams) or local (you, this repo only). Pass --scope from the terminal to skip the prompt. Type /unity: to confirm that the skills appear.

2. Install the Unity CLI. The CLI docs call it experimental and show a version string of 1.0.0-beta.5. Install options include the install script, brew install --cask unity-cli, winget install Unity.CLI, or apt/dnf. Supported systems are Windows 10 21H1+, macOS 14+, and Ubuntu 22.04+ or RHEL 9 with glibc 2.34+.

3. Add the Pipeline package. With the project open, follow the Pipeline package docs:

   unity auth login
   unity pipeline install
   # wait for the Editor to recompile
   unity pipeline list
   

   The package controls the Editor through a local HTTP API and requires Unity 6.0+. Unity's Meet the Unity CLI post calls it experimental and says it will change.

4. Keep the Editor open and run unity status to confirm the CLI can reach it.

What this unlocks

According to the Unity CLI post, unity command eval runs arbitrary C# inside the running Editor. Unity says the code is compiled with Roslyn, runs on the main thread, needs no recompile, and is gated behind a security token. The Pipeline API is localhost-only. A runtime component exists but is off by default and meant for dev builds only.

That is a lot of power to hand an agent. Anyone running it on a real project should commit to source control first.

What limits does Unity admit in its own docs?

Unity's documentation is more candid than its launch post:

• Simple uGUI work: the Claude Code page says the plugin "makes little difference" compared with doing it by hand.
• Sonnet 5: the same page says the plugin mainly supplies correctness. Sonnet 5 builds most Unity features without it but defaults to older approaches.
• Clarifying questions: the About page says the agent asks questions before making changes. That helps in interactive sessions but can block automated pipelines, so pipeline prompts need enough detail to avoid follow-ups.
• Skills don't always trigger from natural phrasing. Name the skill explicitly. If a result looks wrong, the agent may have skipped the references/ folder, so ask it to read those files and retry.
• Updates aren't automatic by default. The blog says new skills show up "on the day we ship it." The docs say Claude Code's auto-update is off by default for third-party marketplaces. To refresh manually, run /plugin marketplace update unity-agent-plugin.
• Missing skills after install: clear ~/.claude/plugins/cache, restart, and reinstall.

One open question: a forum user asked whether manually installed Unity skills conflict with the plugin. The thread shows no direct answer from Unity.

Official skills or community Unity MCP bridges?

Unity's case for its own plugin, per the blog, is that the skills are written by the teams that own each feature, have had a security review, and are documented alongside the engine. The same post credits community plugins for doing good work.

Community bridges take a different route. CoplayDev's unity-mcp describes itself as an MCP bridge that lets AI assistants manage assets, control scenes, edit scripts and automate tasks in the Unity Editor.

The practical difference: Unity's plugin pairs knowledge (skills) with a CLI execution layer, while MCP bridges expose Editor actions as tools. In principle you could run both. None of the sources say whether the two conflict.

Agent tooling is also not universally welcome. One user in the Unity Discussions thread asked for a way to turn off AI promotion in Unity Hub altogether.

Verdict: who should install it?

Install now if you already use Claude Code on a Unity 6+ project and work in areas the skills cover: 2D tilemaps and RuleTiles, UI Toolkit, TextMeshPro CJK fonts, URP Render Graph features, WebGL/WebGPU build size, localization, or Unity IAP and LevelPlay. The skills-only part has the fewest moving parts, and Unity's own docs say its main benefit is steering the model away from outdated approaches.

Wait or skip if:

• you're on Unity 2022 LTS or older (the plugin requires Unity 6.0+)
• your work is mostly simple uGUI, where Unity says it adds little
• you want gameplay code scaffolding, which /new-unity-project explicitly doesn't do
• you run non-interactive CI jobs, where clarifying questions get in the way

Live Editor control is the headline feature, and it rests on two experimental components plus an arbitrary C# eval path. Based on Unity's own labels, expect rough edges until the CLI and Pipeline package leave beta.

Cost: none of the sources list a price for the plugin itself, and the repo is public under the Unity Companion License. You still need a working Claude Code setup.

Follow /lab-notes for updates as the CLI and Pipeline package mature, and check /experiments for step-by-step guides.