qfoldit-unigine-toolbelt
Enables AI agents to automate UNIGINE 2 editor workflows through 25 composite tools (scene, materials, procedural, stamps, codegen, assets, and build console) via an external MCP bridge that runs alongside UNIGINE's official MCPBridge Plugin.
README
qFoldIT Toolbelt — UNIGINE 2 / UNIGINE 2 Sim
25 composite editor-automation tools for UNIGINE 2, exposed to AI agents via a companion MCP bridge that sits alongside UNIGINE's own official MCPBridge Plugin.
Built by qFoldIT — foundation release, 2026
What this is, and how it relates to UNIGINE's MCPBridge Plugin
UNIGINE already publishes an official MCPBridge Plugin on the Add-On Store: a built-in MCP server that runs directly inside the Editor with 27 low-level tools (primitives, node templates, transforms, materials, components, XML inspect/import, console commands). That plugin is the foundation layer — it needs no external process and is the right place for raw editor control.
qFoldIT Toolbelt is a companion, not a replacement. It adds a second layer of higher-level, composite tools the same way UEFN Toolbelt adds 355 named commands on top of UEFN's raw Python API instead of making an agent write one-off scripts every call:
Instead of an agent issuing a dozen raw primitive-creation + transform calls to lay out a competitive arena, it calls:
arena_generate(size="medium")
procedural_place(pattern="circle", count=12, radius=8, node_path="props/crate.node")
material_team_color_split(team_a_contains="RedSpawn", team_b_contains="BlueSpawn")
Why a separate external bridge (and not registered inside MCPBridge itself)
UNIGINE has not currently published a documented API for third parties to
register additional tools directly into MCPBridge's in-process MCP server.
So instead of guessing at an undocumented internal hook, this toolbelt uses
the same two-process pattern UEFN Toolbelt uses for exactly the same
reason (UEFN has no native MCP at all): an external mcp_server.py bridge
that any MCP client connects to over stdio, relaying calls over local HTTP
to a listener running inside the Editor.
Claude / any MCP client
│ MCP protocol (stdio)
▼
mcp_server.py (external, this repo)
│ HTTP POST 127.0.0.1:8766
▼
UNIGINE 2 Editor process
├── MCPBridge Plugin (official, UNIGINE) — 27 base tools, its own MCP server
└── qFoldIT ToolbeltListener.cs (this repo) — 25 composite tools, HTTP listener
Both run side by side. If UNIGINE later documents a plugin-extension API
for MCPBridge, the tool logic here (in editor_plugin/Tools/*.cs) can be
re-registered directly into it — only the transport (mcp_server.py +
ToolbeltListener.cs) would need to change.
See ARCHITECTURE.md for the full design and docs/TOOL_REFERENCE.md for every tool's signature.
⚠ Before you build
editor_plugin/UnigineCompat.cs is the only file that calls
Unigine.* types directly for node/material/transform operations. Its
method bodies reflect the commonly documented UnigineSharp shape for the
2.20/2.21 line (matching what MCPBridge Plugin lists as its supported
SDK), but exact property vs. method access, primitive mesh asset paths,
and the Editor selection API do vary between SDK versions and were not
compiled or run against a live UNIGINE installation to produce this
repository. Check each call against Help → API Documentation in your
SDK Browser before relying on this in a real project — see the comments at
the top of UnigineCompat.cs and StampTools.cs for the specific calls
most likely to need adjustment.
Install
- You already have MCPBridge Plugin (UNIGINE's official plugin) installed — confirmed by your Add-On Store purchase. Keep it enabled; qFoldIT Toolbelt runs alongside it, not instead of it.
- Copy
editor_plugin/into your UNIGINE project (e.g. undersource/qfoldit_toolbelt/) and addToolbeltBootstrapas a world script / autostart WorldLogic component so it initializes with the Editor. - Install the Python bridge dependency:
pip install mcp - Register the bridge with your MCP client. Example for Claude Code
(
.mcp.json):{ "mcpServers": { "qfoldit-unigine-toolbelt": { "command": "python", "args": ["<ABSOLUTE_PATH>/qfoldit-unigine-toolbelt/mcp_server.py"] } } } - Open the Editor (so
ToolbeltBootstrap.Init()starts the listener), then connect your MCP client. Calllist_toolbelt_toolsto confirm the handshake.
Tool categories (25 tools total)
| Category | Tools | What it covers |
|---|---|---|
| Scene | 7 | spawn, transform, clone, delete, parent, list, find nodes |
| Materials | 4 | 12 presets, bulk swap, team-color split, list presets |
| Procedural | 2 | 8-pattern placement, symmetrical arena generator |
| Stamps | 3 | save/place/list reusable node groups |
| Project | 1 | folder scaffold + boilerplate GameManager WorldLogic |
| WorldState | 1 | full world node graph → JSON for AI context |
| CodeGen | 1 | WorldLogic component wired to real world nodes |
| Assets | 3 | list / instantiate / find .node & other project assets |
| BuildConsole | 3 | run console commands, read variables, save world |
Roadmap to parity
This is a foundation release — 25 real tools, not 355 or MCPBridge's
27. Structured to grow the same way UEFN Toolbelt's did: new files under
editor_plugin/Tools/, each calling ToolRegistry.Register(...) for a
handful of new tools, tracked in registry.json.
License
AGPL-3.0, with an additional visible-attribution requirement — see LICENSE. Any tool built on this codebase must credit qFoldIT and link back to this repository (see LICENSE for the exact wording); network/hosted use requires publishing your modified source.
推荐服务器
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。
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。
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。
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。
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。
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。
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。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
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。