threejs-editor-mcp
Enables creating, editing, running, and inspecting small Three.js games inside a DeepSeek Harness chat card via an MCP server and bundled app.
README
threejs-editor-mcp
English | 简体中文
Create, edit, run, and inspect small Three.js games inside a DeepSeek Harness
chat card. One npm package provides the stdio MCP Server, model tools, app-only
tools, and the bundled ui://threejs-editor/app MCP App.

Architecture
flowchart LR
Agent["Harness Agent"] -->|"model-visible tools"| Host["dsh-mcp-apps"]
Host --> Server["threejs-editor-mcp<br/>stdio server"]
Server --> Store["project.json<br/>diagnostics.json<br/>assets/"]
Server -->|"resources/read"| App["MCP App Sandbox<br/>Three.js Editor"]
App -->|"app-only tools"| Host
Human["Human"] --> App
The Editor uses the official three npm package. It is a product-owned,
lightweight editor inspired by the official Three.js Editor; no Editor source
is vendored and no separate web service is required.
Install and Run with npm
Install the MCP Apps Host and this MCP Server:
dsh plugin --profile web add @creative-dswork/dsh-mcp-apps
npm install --global threejs-editor-mcp@0.1.0
threejs-editor-mcp --root /absolute/path/to/threejs-projects
Or run the Server without a global install:
npx --yes threejs-editor-mcp@0.1.0 --root /absolute/path/to/threejs-projects
For a global install, configure the mcp-apps row in the Harness Web profile:
- id: mcp-apps
config:
maxBodyBytes: 2097152
servers:
- serverName: threejs
transport: stdio
command: threejs-editor-mcp
args:
- --root
- /absolute/path/to/threejs-projects
To let npm resolve the package directly, use npx in the same configuration:
- id: mcp-apps
config:
maxBodyBytes: 2097152
servers:
- serverName: threejs
transport: stdio
command: npx
args:
- --yes
- threejs-editor-mcp@0.1.0
- --root
- /absolute/path/to/threejs-projects
The root can also be provided with THREEJS_EDITOR_PROJECT_ROOT.
Capabilities
emptyand playablepongproject templates;- hierarchy, viewport picking, transform controls, Inspector, undo/redo;
- one
start/update/disposegame script with keyboard and pointer input; - revision-checked human saves and model changes;
- revision-bound Play diagnostics;
- GLB, PNG, and JPEG import plus self-contained project export;
- clean reload and dirty conflict preservation.
Model-visible tools:
list_projects, create_project, open_editor, inspect_project,
apply_scene_changes, and check_project.
App-only tools:
pull_project, push_project, save_project_copy, report_diagnostics,
put_asset, and export_project.
Asset Limits
Assets are local only. Each file is limited to 256 KiB, each project to eight
files and 512 KiB total. The Server validates names, MIME types, signatures,
GLB 2.0 headers, and rejects external GLB resource URIs. Supported extensions
are .glb, .png, .jpg, and .jpeg.
The JSON export contains the saved project, exact revision, and base64 copies of its original assets. The MCP Apps Host mediates the download because direct downloads are blocked inside the Sandbox.
Development
Requires Node.js 22.19 or newer and pnpm 10.25.
pnpm install
pnpm run check
pnpm run test:pack
Run the server directly:
pnpm run build
node dist/server.js --root .tmp/projects
The complete design and phased evidence are in DESIGN.md and
reports/.
Security
Projects and assets are confined to the configured root. Project writes are size-bounded, revision-checked, serialized per project, and committed with temporary-file rename. Project paths, asset paths, symlinks, stale revisions, oversized payloads, mismatched file types, and external asset URLs are rejected. Game scripts run only in the existing different-origin double iframe Sandbox.
License
MIT
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。