Unity Plugins Claude Code Codex 2026 Complete Guide

A Unity coding agent can do more than suggest C# when it has access to task-specific instructions and project tools. For developers, that can mean help with real files and Editor workflows, not just code snippets. It also means a mistake can reach a scene or asset, not only a draft of a script.

A skill is a reusable set of task-specific instructions an AI agent can load to follow a defined workflow. This guide helps you choose an agent, install the official integration, understand what it can change, and protect your project. Hands-on Editor access needs extra care because some actions affect the open project directly.

You will see where the Unity Agent Plugin fits, how its skills differ from a model and its tools, and how supported agents handle access. You will also get practical steps for checking installs and reviewing project changes. The practical starting point is to match the integration to the agent you already use, then keep control of project changes.

Key Takeaways

• The Unity Agent Plugin adds Unity guidance for supported coding agents.
• Unity 6 and later is the baseline for the plugin.
• Skills guide workflows, but they do not guarantee correct results.
• Claude Code, Codex, Grok and Antigravity support the plugin in 2026.
• Version control gives you a stronger recovery path than Editor undo alone.

Unity Plugins Claude Code Codex 2026: What the Official Integration Includes

Unity’s official integration packages engine-specific guidance for two major coding agents, with Unity 6 and later as its project baseline. The Unity Agent Plugin adds task instructions and supported workflows, rather than introducing a new AI model. That distinction matters when you compare it with the coding tool you already use.

The Unity Agent Plugin supports Unity 6 and later. A project using Unity 2022 falls outside that stated baseline. The plugin gives a supported agent Unity-focused guidance, while the agent and its model still handle the conversation and coding work.

Check the project’s Unity version before choosing an installation path. The official plugin’s baseline is Unity 6 and later.

The first Claude Code plugin launched on September 9, 2026, with 29 skills. The Codex plugin launched on September 16, 2026, with 31 skills. Those release counts give you a sense of the initial task coverage, but they do not mean every skill performs the same actions.

The plugin is an integration layer between Unity’s guidance and a coding agent. It does not replace the model that reads your request, and it does not make every agent feature available in every client. The project version remains the first practical detail to check.

What Are Unity Skills, and How Do Coding Agents Use Them?

A skill is a reusable set of task-specific instructions an agent can load when a job calls for it. Unity skills give Claude Code and Codex guidance for relevant work, rather than adding a new model. Think of them as a work pattern, not as a new source of intelligence.

Tools enable actions, while skills guide how and when an agent should use them. A model interprets your request and reasons about possible steps. Tools can let the agent inspect files, make edits, or run supported operations.

A Unity skill can prompt the agent to check project context before editing. It can direct the agent toward current workflows and warn about common mistakes. It can also ask for validation steps after a change.

That guidance can help an agent follow Unity-specific documentation instead of relying only on general coding patterns. For example, a skill for a Unity system can give the agent task-specific steps that fit that system. The model still has to apply those steps to your actual project.

Skills do not give an agent abilities it otherwise lacks. A written workflow cannot open an Editor or change an asset unless the client has suitable tools and access. The available actions depend on the tools connected to that agent.

Results can still be wrong, even when the agent loads the right skill and completes each requested step. A skill guides the process, but it cannot confirm that your game behaves as intended without a meaningful check. Treat a skill as workflow guidance, not as a substitute for reviewing the agent’s changes.

Which Unity Tasks Do the Skills Cover?

The catalog spans both everyday project work, such as UI and 2D setup, and more specialized systems such as monetization. Its named skills can help you match a task to relevant Unity guidance. The task still sets the scope, since a skill does not promise one level of automation for every project.

Use the catalog to narrow a request to a specific Unity work area. These examples show where the guidance can apply and the kind of task you might give an agent.

UI, rendering and 2D workflows

For interface work, UI Toolkit and uGUI cover different Unity UI workflows. You can ask an agent to adjust a screen or build an interface element, then inspect the result in the project. The skill can guide its steps, but the task should name the intended screen and behavior.

Rendering guidance includes URP, the Universal Render Pipeline, and Shader Graph. These examples suit work on a render pipeline or a shader, where project settings and existing assets matter. For a 2D development task, sprites and tilemaps can guide work on level setup and tile rules.

Multiplayer, audio and platform systems

Audio, navigation, physics and multiplayer skills cover common systems that shape gameplay. Their work can cross scripts, scene setup and project configuration. A request about a single behavior, such as how a character moves through a level, gives the agent a clearer target than “fix gameplay.”

Localization and web builds can support platform preparation. They point toward tasks that involve language settings or build targets, rather than a general promise to ship a game. Review the project settings and output that the task touches.

Monetization and live-game operations

Unity IAP and LevelPlay are the named monetization examples. They can help frame tasks around in-app purchases or advertising workflows. Those systems affect how players spend or see ads, so the desired player-facing result should be clear.

The catalog connects skills to specific areas, from UI Toolkit screens to Unity IAP. It does not mean that every task runs without input or review. Match the skill’s stated task to the change you actually want, rather than asking it to handle a loosely defined project goal.

How Do the Unity CLI and Editor Control Work?

The Unity CLI connects agent workflows to Unity operations that go beyond editing a source file. It is a command-line tool for Editor installation, project creation and opening, package management, builds, tests and logs. That wider reach can connect a coding task to Unity operations.

When the Unity CLI is available, supported skills can run C# in the open Unity Editor. A workflow may use the CLI to reach that Editor and make changes in the project. That differs from an agent editing a standalone C# file without a live Editor connection.

Unity Hub installs the Unity CLI automatically, so you may already have it if you use Hub. The CLI can also help with tasks that need Unity operations, such as starting a build or checking logs. The exact work depends on the skill and the tools available to the agent.

A live Editor workflow can affect more than the source code that appears in a diff. The agent may interact with an open project and its current state. For that reason, know which project is open before using a task that reaches the Editor.

The Unity CLI is a tool, not a separate AI model. It supplies a route to Unity operations that a coding agent can use when the workflow supports it. Before relying on live Editor work, confirm that the Unity project is open and the CLI is available.

How to Install the Unity Plugin in Claude Code

For the marketplace route, run the two plugin commands from inside a Claude Code session, in order. Type the slash commands in Claude Code, not in a shell. The first command adds Unity’s marketplace, and the second installs the plugin.

Use the in-session commands in this order:

1. Type /plugin marketplace add Unity-Technologies/unity-agent-plugin inside Claude Code.
2. Type /plugin install unity@unity-agent-plugin inside the same session.

The terminal route uses a different command form. Open a terminal and run claude plugin marketplace add Unity-Technologies/unity-agent-plugin. Then run claude plugin install unity@unity-agent-plugin. These commands belong in your shell, unlike the slash commands above.

A terminal install loads in the next Claude Code session. If the session is already open, you can use /reload-plugins instead of closing it. Keep the command forms straight: typing a terminal command in the Claude Code prompt will not install the plugin.

The plugin adds Unity-specific guidance to Claude Code, but it does not change the project’s Unity version requirement. Use it with a Unity 6 or later project. After a terminal install, start a new session or run /reload-plugins in the open session to load the plugin.

How to Install the Plugin in Codex, Grok and Antigravity

Codex uses two terminal commands to add Unity’s marketplace and install its plugin. Grok Build and Antigravity offer different installation surfaces, so each client needs its own path. Do not swap commands between those tools.

Choose the commands or app menu that match the client you have open.

Add the plugin to Codex

Open a terminal and run the marketplace command first. Then add the Unity plugin through that marketplace.

1. Run codex plugin marketplace add Unity-Technologies/unity-agent-plugin.
2. Run codex plugin add unity@unity-agent-plugin.

The order matters because Codex must know the marketplace before it can add the plugin from that marketplace. These are terminal commands, not commands to type into another agent’s prompt.

Install through Grok Build or Antigravity

In Grok Build, open /marketplace and find Unity’s plugin. Install it from the marketplace interface. Grok Build also supports a terminal install with grok plugin install Unity-Technologies/unity-agent-plugin –trust.

Antigravity has an app-based route under Customizations > Plugins. Find the Unity plugin there and install it. From a terminal, use agy plugin install https://github.com/Unity-Technologies/unity-agent-plugin.

These clients do not share one installation surface. Codex uses marketplace commands, Grok Build has its own marketplace and terminal path, and Antigravity has both app and terminal options. Use the installation path that matches the client you have open, and check that the install completed before starting a task.

Which Supported Unity Agent Fits Your Existing Workflow?

All four clients have an official Unity plugin, so your existing agent preference can guide the choice. The Unity Agent Plugin is available for Claude Code, Codex, Grok and Antigravity in 2026. The main difference here is how each client exposes or verifies its skills.

You can keep your preferred coding agent, then use its own interface to find the Unity skills. Teams do not need to switch tools just to use the official integration.

Claude Code and its skill menu

Claude Code exposes Unity skills through the /unity: prefix. Type that prefix to see matching skill commands in the menu. You can also check /plugin to confirm that the Unity plugin is enabled.

That interface can suit developers who already work in Claude Code and want to call Unity-specific workflows from its skill menu. If the prefix does not appear, check the plugin status before treating a skill as unavailable. The menu is the skill access point, not a sign that the result is correct.

Codex, Grok and Antigravity access

Codex provides codex plugin list to check plugin status. The listed details show version 0.1.8-beta for Codex. Grok exposes skills in its slash menu; skill names appear in plain form, with qualified names available when a name conflicts.

Antigravity also exposes skills through a slash menu, and agy plugin list can show its plugin status. The listed Antigravity version is 0.1.8-beta. These access differences matter more than the agent label if your team already has a familiar way to review tasks.

For broader agent selection, compare how AI coding assistants and agents fit your work across repository tasks and review needs. Verify the plugin in the client you intend to use before relying on its skills in a project.

What Does Claude Code Cost in 2026?

The relevant cost is the Claude plan that provides Claude Code access, not a price claim about the Unity plugin. Claude Code is included with Claude Pro and Claude Max, but not Claude Free in 2026. Plan access sets the cost, while usage limits still apply.

Claude Pro costs $20 monthly or $17 per month with annual billing. The annual option is billed as $200 up front. Claude Max has two usage levels with higher monthly prices.

Individual Claude plans

Claude Pro costs $20 per month when billed monthly, or $17 per month with annual billing. The annual price is $200 billed up front. Claude Max costs $100 per month for 5x Pro usage or $200 per month for 20x Pro usage.

Those figures describe Claude plan access, not the Unity Agent Plugin. Plan usage limits apply, so a higher tier changes the available usage level rather than removing all limits. You can read a Claude Code plan price comparison alongside the included access details.

Before choosing a plan, think about how often you use Claude Code and how much usage your work needs. A small Unity task and daily coding work may place different demands on a plan. Check what the usage level means for your own routine.

Claude plans for teams and enterprises

Claude Team standard costs $25 per seat each month with monthly billing, or $20 per seat per month with annual billing. Team premium costs $125 per seat per month monthly, or $100 per seat per month annually. A premium seat includes 5x the usage of a standard seat.

Claude Enterprise self-serve costs US$20 per seat per month with annual billing and a 20-seat minimum. Usage is charged at API rates. These terms make Enterprise different from a simple per-seat plan with included Claude Code usage.

Claude plan prices may change at Anthropic’s discretion, and usage limits apply across plans. Check the current plan terms before choosing a subscription, because usage limits apply and prices can change.

How to Protect a Unity Project Before Agent Changes

Commit the project before asking an agent to make a large set of changes. Skills can change scripts, scenes and assets directly, so a clean recovery point matters. A commit gives you a known state to compare with the agent’s work.

Editor undo is not a complete recovery plan. Some changes may not be logged or undoable from the Editor, especially when a workflow affects several project files or assets.

• Keep the Unity project in version control.
• Commit before a large agent-driven task.
• Inspect the changes the agent made.
• Test the project before accepting the work.

Version control lets you track files and manage changes across a project. Unity’s version control documentation covers Editor integrations and file-change workflows. A commit also gives you a clear point for reviewing what changed after an agent task.

Inspect scripts, scenes and assets rather than judging only by the agent’s summary. A task can finish without errors while still changing the wrong object or behavior. Testing in the project helps reveal issues that a text diff cannot show.

For small edits, use judgment about how much review the task needs, but keep the project under version control. Treat the version-control history as the recovery point, not the Editor’s undo stack.

How to Check, Update or Manually Add Unity Skills

Check the client’s plugin list first, especially if a skill command does not appear. Codex and Antigravity list the Unity Agent Plugin as version 0.1.8-beta in 2026. A missing command can be an install issue, not a problem with the task itself.

Use codex plugin list for Codex and agy plugin list for Antigravity. If marketplace installation is not available, you can link skill folders from a cloned repository into your personal skills directory.

The manual route keeps the skill files in a local clone and points the agent to them. It is a fallback for Claude Code or Codex, not the same as a marketplace plugin install. Linked skills load in later sessions and do not appear under a unity: plugin prefix.

1. Clone Unity-Technologies/unity-agent-plugin.
2. Create the personal skills directory for your client, such as ~/.claude/skills or ~/.agents/skills.
3. Link each skill folder into the appropriate directory for Claude Code or Codex.
4. Run git pull in the clone when you want to update the linked skills.

Claude Code uses ~/.claude/skills, while Codex uses ~/.agents/skills for this manual path. Link the skill folders themselves into the matching directory. The repository clone remains the place where you fetch updated skill files.

Manually linked skills appear in later sessions, but they do not use the plugin prefix in the command menu. That difference can help you tell a linked skill from one loaded through the official plugin. If you rely on linked skills, run git pull in the cloned repository when you want to update them.

How Does the Unity MCP Server Differ From the Plugin?

The plugin packages Unity guidance for an agent, while the MCP Server exposes live Editor information and actions to a connected client. Unity described the Unity MCP Server in May 2026 as part of the open-beta Unity AI Assistant package. One adds guidance; the other adds an Editor connection.

The Unity MCP Server exposes project context and Editor operations to connected agents. Its setup depends on Unity services and client approval, rather than an agent-specific plugin install alone.

What the MCP Server exposes

The MCP Server can expose scene hierarchies, GameObjects, component values, build settings and console messages. Its tools include scene management, C# script editing, console access, GameObject component inspection and build-settings inspection.

You can also register custom MCP tools in C#. That gives teams a way to add project-specific actions to the connection. The MCP Server’s tools are different from the task guidance inside a skill, even when both support work in the same project.

Requirements and first-connection approval

The MCP Server requires Unity 6 (6000.0) or later and the AI Assistant package. The project must connect to Unity Cloud, and you need an active Unity AI tools beta trial or subscription. All four conditions matter before you troubleshoot a client.

Approve a client’s first connection in Edit > Project Settings > AI > Unity MCP. Previously approved clients reconnect automatically. When Unity starts, it installs the MCP relay at ~/.unity/relay/.

Manual client configuration passes –mcp to the relay executable. The client must connect through that relay to reach the Editor features. Confirm the project, package, cloud connection and beta access before troubleshooting a client connection.

How to Get More Reliable Results From a Unity Coding Agent

A focused request gives the agent a clearer task than asking it to improve a whole game project. Name the behavior or asset you want changed and add the relevant project context. Specific requests leave less room for the agent to guess.

Give the coding agent one task with a result you can observe. A suitable skill can guide the workflow, but successful execution does not prove that the result is correct.

Check the change against the behavior you wanted, not just the agent’s report that the task finished.

For in-app purchases, ask the agent to use the implement-in-app-purchases skill and name the purchase flow you need. That gives it a defined feature to work on instead of a broad request to “add monetization.”

When building a tile rule from existing terrain or edge sprites, request tilemap-ruletile-createfromsegment. Tell the agent which sprites already belong to the level and what connections the tile should support. Then check whether the rule fits those assets.

For a text issue, describe the affected label and ask for help with missing Chinese characters in TextMeshPro. Mention where the label appears and which characters fail to display. You can then inspect the same label in the game or Editor.

A rendering task can be just as bounded. Ask the agent to review a ScriptableRendererFeature for Render Graph problems, and point it to the relevant code or project area. That request targets a known part of the render workflow.

For more context on task fit, see how code editors and AI coding workflows compare across different developer needs. The details you give should help the agent choose a relevant workflow, while your own check should confirm the actual behavior.

Give the agent one observable outcome to verify, such as a correctly formed tile rule or readable text in the target label.

Conclusion

The useful question is whether Unity’s agent guidance fits the work you already do, not whether an agent should make every project decision. Claude Code, Codex, Grok and Antigravity give you different ways to use the official integration. Pick the client that fits your current workflow.

Direct project edits still need review, especially when a task reaches scenes, assets or the open Editor. Start with one small, reversible task in a Unity 6 or later project after committing the project. Review every change before expanding the workflow.