Agent Bridge for Unity

Agent Bridge for Unity

Enables AI agents to control the Unity Editor through MCP, allowing scene building, runtime scripting, visual QA, and more.

Category
访问服务器

README

Agent Bridge for Unity

License: Apache 2.0 Unity 6+ .NET 10

Let AI actually operate the Unity Editor — build scenes, run Play Mode, capture screenshots, and iterate from visual feedback.

Works with Claude Code, Cursor, Windsurf, VS Code, or any MCP-compatible client.

Scene Building

Not just code edits — the agent can create objects, adjust lighting, wait for compilation, enter Play Mode, take screenshots, compare against references, and keep refining until the scene is right.


What Can It Do?

Runtime Scripting — Create C# scripts, wait for compilation, attach behaviors, enter Play Mode, and validate results.

Runtime Scripting

Visual QA — Adjust lighting and materials, capture screenshots, compare against references, refine iteratively.

Visual QA

Scene Reconstruction — Reconstruct 3D structures from reference images using iterative visual validation.

Scene Reconstruction

Try prompts like:

Create a stylized forest with 30 trees, scattered rocks, and warm lighting.

Block out a 2D platformer level with floating platforms, a player spawn, and a goal area.

Set up horror lighting: dim blue ambient, flickering point light, red emergency light, volumetric fog.

Build an obstacle course with moving platforms, a rotating blade, and a jump pad.
Then add a player controller and a win condition.

Create a solar system with orbiting planets, add the scripts, enter Play Mode, and verify it works.

Quick Start

Prerequisites

  • Unity 6 (6000.0+) with URP
  • .NET 10 SDK
  • An MCP-compatible client

1. Install the Package

Git URL (recommended) — In Unity: Window > Package Manager > + > Add package from git URL:

https://github.com/sebastiankurvers-dev/agent-bridge-for-unity.git

<details> <summary>Other install methods</summary>

Local clone:

cd YourUnityProject/Packages
git clone https://github.com/sebastiankurvers-dev/agent-bridge-for-unity.git com.sebastiankurvers.agent-bridge

Local path reference — add to Packages/manifest.json:

{
  "dependencies": {
    "com.sebastiankurvers.agent-bridge": "file:/path/to/agent-bridge-for-unity"
  }
}

</details>

2. Verify the Bridge

The bridge starts automatically on port 5847 when Unity compiles. Check Window > Agent Bridge in Unity, or:

curl http://127.0.0.1:5847/health

3. Configure Your MCP Client

Create .mcp.json in your Unity project root:

{
  "mcpServers": {
    "unity": {
      "type": "stdio",
      "command": "dotnet",
      "args": ["run", "--project", "Packages/com.sebastiankurvers.agent-bridge/UnityMCP~/UnityMCP.csproj"],
      "env": {}
    }
  }
}

If using a local path install, replace with the absolute path to your UnityMCP.csproj.

4. Start Using It

Open your MCP client from the Unity project directory. Tools are available immediately.


How It Works

┌─────────────────┐       ┌──────────────────────────────┐
│   MCP Client    │       │  UnityMCP (.NET 10 Server)   │
│  (Claude Code,  │─stdio─│  Receives tool calls,        │
│  Cursor, etc.)  │       │  forwards to Unity over HTTP │
└─────────────────┘       └──────────────┬───────────────┘
                                         │
                                    HTTP :5847
                                         │
                          ┌──────────────┴───────────────┐
                          │  UnityAgentBridge (Editor)    │
                          │  Dispatches to Unity main     │
                          │  thread, executes commands    │
                          └──────────────┬───────────────┘
                                         │
                                  Unity Editor API
Layer Location Role
UnityMCP UnityMCP~/ .NET 10 MCP server. Receives tool calls via stdio, forwards to Unity over HTTP.
UnityAgentBridge Editor/UnityAgentBridge/ Unity Editor plugin. HTTP server on 127.0.0.1:5847, dispatches to main thread.

The ~ suffix on UnityMCP~/ tells Unity to ignore the folder — it won't compile .NET 10 code as Unity scripts.


Tool Coverage

248 tools across scene building, terrain, particles, visual QA, automation, editor control, assets, and scripting. You don't need to learn them — your AI agent discovers and calls them as needed.

<details> <summary>Full tool breakdown</summary>

Category Tools Description
Scene Builder 30 Spawn prefabs, primitives, path-based placement, descriptors, transactions
Assets 21 Find, search, catalog, import settings, geometry metadata
GameObjects 20 Inspect, modify, delete, reparent, group, scatter, batch operations
Rendering 19 Camera settings, render pipeline, material preview, shader scoping
Scene 18 Save, load, hierarchy queries, scene profiling
UI Toolkit 14 UXML/USS creation, visual element queries, binding
Components 12 Add, remove, inspect, modify, patch serialized properties
Screenshots 12 Capture, compare with suggestions, multi-POV, frame sequences
Lighting 9 Lights, render settings, volumes, procedural skybox, reflection probes
Camera 9 Scene view control, orbit, focus, screenshot viewpoints
Animation 8 Animator controllers, states, transitions, FBX clips
Shaders 7 Create, inspect, keywords, property management
Scripts 7 Create, modify, read structure, type schemas
Terrain 6 Create, sculpt, paint texture layers, place trees
Replay 5 Record, execute, compare — full state verification
Checkpoints 5 Save/restore scene snapshots for rollback
Spatial Audit 7 Overlap detection, grounding audit, snap-to-ground, object identification at screen points
Physics 5 Rigidbody, colliders, physics settings, spatial queries
Audio 5 Sources, mixers, listeners, spatial audio
Play Mode 5 Enter/exit, wait, event polling
Primitives 4 Built-in primitives, procedural meshes, raw mesh API, compound shapes
Performance 4 Telemetry, baselines, profiling
Packages 4 List, add, remove, search Unity packages
Particles 3 Preset templates, get/configure particle systems
Tests 2 Run tests, get results
Console 2 Read and clear Unity console logs
Workflow Guide 1 Task-specific tool recommendations

</details>


Compatibility

Requirement Supported
Unity 6000.0+ (Unity 6)
Render Pipeline URP
.NET SDK 10.0+
OS Windows, macOS (Intel & Apple Silicon), Linux

Security

The bridge listens on 127.0.0.1 only — not accessible from other machines.

No authentication required by default. For shared machines, set BRIDGE_AUTH_TOKEN:

export BRIDGE_AUTH_TOKEN="your-secret-token"

<details> <summary>Additional security options</summary>

  • C# execution: Enabled by default with sandbox restrictions. Disable with BRIDGE_DISABLE_EXECUTE=1
  • Body size limit: 10 MB default (BRIDGE_MAX_REQUEST_BODY_BYTES)
  • Concurrency limit: 32 concurrent requests (BRIDGE_MAX_CONCURRENT_REQUESTS)
  • Route allowlist: BRIDGE_ALLOWLIST_FILE="/path/to/allowlist.json"
  • Audit logging: BRIDGE_AUDIT_LOG="/path/to/audit.log"

</details>


Troubleshooting

Problem Fix
Bridge not starting Check Window > Agent Bridge in Unity. Ensure port 5847 is free.
MCP server won't build Verify .NET 10: dotnet --version. Try dotnet build UnityMCP~/UnityMCP.csproj
Tools not appearing Ensure .mcp.json is in project root. Restart your MCP client.
Scene save dialogs Intentional — the bridge auto-saves before write operations to prevent modal dialogs.

Support

If this project saves you time, consider buying me a coffee.

Community

Contributing | Security Policy | Code of Conduct

License

Apache 2.0

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选