React Flow架构指南
react-flow-architecture
by anderskev
Architectural guidance for building node-based UIs with React Flow. Use when designing flow-based applications, making decisions about state management, integration patterns, or evaluating whether React Flow fits a use case.
安装
claude skill add --url https://github.com/openclaw/skills文档
React Flow Architecture
When to Use React Flow
Good Fit
- Visual programming interfaces
- Workflow builders and automation tools
- Diagram editors (flowcharts, org charts)
- Data pipeline visualization
- Mind mapping tools
- Node-based audio/video editors
- Decision tree builders
- State machine designers
Consider Alternatives
- Simple static diagrams (use SVG or canvas directly)
- Heavy real-time collaboration (may need custom sync layer)
- 3D visualizations (use Three.js, react-three-fiber)
- Graph analysis with 10k+ nodes (use WebGL-based solutions like Sigma.js)
Architecture Patterns
Package Structure (xyflow)
@xyflow/system (vanilla TypeScript)
├── Core algorithms (edge paths, bounds, viewport)
├── xypanzoom (d3-based pan/zoom)
├── xydrag, xyhandle, xyminimap, xyresizer
└── Shared types
@xyflow/react (depends on @xyflow/system)
├── React components and hooks
├── Zustand store for state management
└── Framework-specific integrations
@xyflow/svelte (depends on @xyflow/system)
└── Svelte components and stores
Implication: Core logic is framework-agnostic. When contributing or debugging, check if issue is in @xyflow/system or framework-specific package.
State Management Approaches
1. Local State (Simple Apps)
// useNodesState/useEdgesState for prototyping
const [nodes, setNodes, onNodesChange] = useNodesState(initialNodes);
const [edges, setEdges, onEdgesChange] = useEdgesState(initialEdges);
Pros: Simple, minimal boilerplate Cons: State isolated to component tree
2. External Store (Production)
// Zustand store example
import { create } from 'zustand';
interface FlowStore {
nodes: Node[];
edges: Edge[];
setNodes: (nodes: Node[]) => void;
onNodesChange: OnNodesChange;
}
const useFlowStore = create<FlowStore>((set, get) => ({
nodes: initialNodes,
edges: initialEdges,
setNodes: (nodes) => set({ nodes }),
onNodesChange: (changes) => {
set({ nodes: applyNodeChanges(changes, get().nodes) });
},
}));
// In component
function Flow() {
const { nodes, edges, onNodesChange } = useFlowStore();
return <ReactFlow nodes={nodes} onNodesChange={onNodesChange} />;
}
Pros: State accessible anywhere, easier persistence/sync Cons: More setup, need careful selector optimization
3. Redux/Other State Libraries
// Connect via selectors
const nodes = useSelector(selectNodes);
const dispatch = useDispatch();
const onNodesChange = useCallback((changes: NodeChange[]) => {
dispatch(nodesChanged(changes));
}, [dispatch]);
Data Flow Architecture
User Input → Change Event → Reducer/Handler → State Update → Re-render
↓
[Drag node] → onNodesChange → applyNodeChanges → setNodes → ReactFlow
↓
[Connect] → onConnect → addEdge → setEdges → ReactFlow
↓
[Delete] → onNodesDelete → deleteElements → setNodes/setEdges → ReactFlow
Sub-Flow Pattern (Nested Nodes)
// Parent node containing child nodes
const nodes = [
{
id: 'group-1',
type: 'group',
position: { x: 0, y: 0 },
style: { width: 300, height: 200 },
},
{
id: 'child-1',
parentId: 'group-1', // Key: parent reference
extent: 'parent', // Key: constrain to parent
position: { x: 10, y: 30 }, // Relative to parent
data: { label: 'Child' },
},
];
Considerations:
- Use
extent: 'parent'to constrain dragging - Use
expandParent: trueto auto-expand parent - Parent z-index affects child rendering order
Viewport Persistence
// Save viewport state
const { toObject, setViewport } = useReactFlow();
const handleSave = () => {
const flow = toObject();
// flow.nodes, flow.edges, flow.viewport
localStorage.setItem('flow', JSON.stringify(flow));
};
const handleRestore = () => {
const flow = JSON.parse(localStorage.getItem('flow'));
setNodes(flow.nodes);
setEdges(flow.edges);
setViewport(flow.viewport);
};
Integration Patterns
With Backend/API
// Load from API
useEffect(() => {
fetch('/api/flow')
.then(r => r.json())
.then(({ nodes, edges }) => {
setNodes(nodes);
setEdges(edges);
});
}, []);
// Debounced auto-save
const debouncedSave = useMemo(
() => debounce((nodes, edges) => {
fetch('/api/flow', {
method: 'POST',
body: JSON.stringify({ nodes, edges }),
});
}, 1000),
[]
);
useEffect(() => {
debouncedSave(nodes, edges);
}, [nodes, edges]);
With Layout Algorithms
import dagre from 'dagre';
function getLayoutedElements(nodes: Node[], edges: Edge[]) {
const g = new dagre.graphlib.Graph();
g.setGraph({ rankdir: 'TB' });
g.setDefaultEdgeLabel(() => ({}));
nodes.forEach((node) => {
g.setNode(node.id, { width: 150, height: 50 });
});
edges.forEach((edge) => {
g.setEdge(edge.source, edge.target);
});
dagre.layout(g);
return {
nodes: nodes.map((node) => {
const pos = g.node(node.id);
return { ...node, position: { x: pos.x, y: pos.y } };
}),
edges,
};
}
Performance Scaling
Node Count Guidelines
| Nodes | Strategy |
|---|---|
| < 100 | Default settings |
| 100-500 | Enable onlyRenderVisibleElements |
| 500-1000 | Simplify custom nodes, reduce DOM elements |
| > 1000 | Consider virtualization, WebGL alternatives |
Optimization Techniques
<ReactFlow
// Only render nodes/edges in viewport
onlyRenderVisibleElements={true}
// Reduce node border radius (improves intersect calculations)
nodeExtent={[[-1000, -1000], [1000, 1000]]}
// Disable features not needed
elementsSelectable={false}
panOnDrag={false}
zoomOnScroll={false}
/>
Trade-offs
Controlled vs Uncontrolled
| Controlled | Uncontrolled |
|---|---|
| More boilerplate | Less code |
| Full state control | Internal state |
| Easy persistence | Need toObject() |
| Better for complex apps | Good for prototypes |
Connection Modes
| Strict (default) | Loose |
|---|---|
| Source → Target only | Any handle → any handle |
| Predictable behavior | More flexible |
| Use for data flows | Use for diagrams |
<ReactFlow connectionMode={ConnectionMode.Loose} />
Edge Rendering
| Default edges | Custom edges |
|---|---|
| Fast rendering | More control |
| Limited styling | Any SVG/HTML |
| Simple use cases | Complex labels |
相关 Skills
前端设计
by anthropics
面向组件、页面、海报和 Web 应用开发,按鲜明视觉方向生成可直接落地的前端代码与高质感 UI,适合做 landing page、Dashboard 或美化现有界面,避开千篇一律的 AI 审美。
✎ 想把页面做得既能上线又有设计感,就用前端设计:组件到整站都能产出,难得的是能避开千篇一律的 AI 味。
网页构建器
by anthropics
面向复杂 claude.ai HTML artifact 开发,快速初始化 React + Tailwind CSS + shadcn/ui 项目并打包为单文件 HTML,适合需要状态管理、路由或多组件交互的页面。
✎ 在 claude.ai 里做复杂网页 Artifact 很省心,多组件、状态和路由都能顺手搭起来,React、Tailwind 与 shadcn/ui 组合效率高、成品也更精致。
网页应用测试
by anthropics
用 Playwright 为本地 Web 应用编写自动化测试,支持启动开发服务器、校验前端交互、排查 UI 异常、抓取截图与浏览器日志,适合调试动态页面和回归验证。
✎ 借助 Playwright 一站式验证本地 Web 应用前端功能,调 UI 时还能同步查看日志和截图,定位问题更快。
相关 MCP 服务
GitHub
编辑精选by GitHub
GitHub 是 MCP 官方参考服务器,让 Claude 直接读写你的代码仓库和 Issues。
✎ 这个参考服务器解决了开发者想让 AI 安全访问 GitHub 数据的问题,适合需要自动化代码审查或 Issue 管理的团队。但注意它只是参考实现,生产环境得自己加固安全。
Context7 文档查询
编辑精选by Context7
Context7 是实时拉取最新文档和代码示例的智能助手,让你告别过时资料。
✎ 它能解决开发者查找文档时信息滞后的问题,特别适合快速上手新库或跟进更新。不过,依赖外部源可能导致偶尔的数据延迟,建议结合官方文档使用。
by tldraw
tldraw 是让 AI 助手直接在无限画布上绘图和协作的 MCP 服务器。
✎ 这解决了 AI 只能输出文本、无法视觉化协作的痛点——想象让 Claude 帮你画流程图或白板讨论。最适合需要快速原型设计或头脑风暴的开发者。不过,目前它只是个基础连接器,你得自己搭建画布应用才能发挥全部潜力。