sysdesign
Enables Claude to design system architectures and scaffold minimal Python/Docker implementations, with live D3 visualization in the browser.
README
MCP System Design — live architecture + scaffolding
A web app you drive through Claude over MCP. Prompt Claude to design a system → the browser draws the architecture live (interactive D3 nodes & edges). Say "develop" → Claude scaffolds a minimal Python + Docker MVP into a folder.
┌─────────┐ MCP (stdio) ┌───────────────────────┐ SSE ┌─────────┐
│ Claude │ ───────────────▶ │ server.py │ ───────▶ │ Browser │
│ (client)│ visualize_... │ MCP tools + FastAPI │ /events │ D3 UI │
└─────────┘ scaffold_... │ + shared state │ └─────────┘
└───────────────────────┘
│ writes
▼
generated/<project>/ (FastAPI services + Docker)
What's inside
| File | Role |
|---|---|
server.py |
MCP server (stdio) + embedded FastAPI web server + SSE endpoint |
scaffolder.py |
Turns a design into FastAPI services + Dockerfiles + docker-compose |
state.py |
Thread-safe shared state watched by the SSE stream |
web/index.html |
D3 force-graph UI, live via EventSource("/events") |
MCP tools Claude can call
visualize_architecture(title, components, connections)— draw the diagram.components:[{id, label, type?, description?}](typecolors the node:service,queue,db/store,sse,cache,gateway).connections:[{source, target, label?}].scaffold_implementation(project_name, base_port=8001)— generate the MVP from whatever is currently on screen. Writes togenerated/<project_name>/.get_current_design()— return the design currently displayed.
Run it
1. Start the web UI
pip install -r requirements.txt
APP_MODE=web python server.py # UI at http://localhost:8765
or with Docker:
docker compose up --build # UI at http://localhost:8765
2. Connect Claude to the MCP server
Add to your MCP client config (e.g. Claude Desktop claude_desktop_config.json).
This launches the server over stdio; the web UI comes up automatically on 8765.
{
"mcpServers": {
"sysdesign": {
"command": "python",
"args": ["/absolute/path/to/mcp-sysdesign/server.py"]
}
}
}
Note: the MCP client owns stdio, so let it launch
server.py(defaultAPP_MODE=mcpruns MCP and the web UI). Don't also runAPP_MODE=webon the same port, or 8765 will be taken.
3. Use it
Open http://localhost:8765, then in Claude:
"Design a URL crawler system that stores page text keyed by URL, with an SSE monitoring channel."
Claude calls visualize_architecture(...) and the graph appears. Then:
"develop"
Claude calls scaffold_implementation("crawler") and generated/crawler/
appears with runnable services:
cd generated/crawler && docker compose up --build
Notes / MVP scope
- SSE (not WebSocket) powers the live UI: it's a one-way server→browser push, which is exactly what a live diagram needs, and it auto-reconnects.
- The scaffold is a skeleton — each service builds, starts, and can reach its declared dependencies over the Docker network. Business logic is left as TODO.
- State is in-memory (single process). Restarting the server clears the diagram.
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。