drawio-mcp
Enables AI assistants to create, read, and edit draw.io diagrams with live preview in VS Code.
README
drawio-mcp
MCP server that lets AI assistants create, read, and edit draw.io diagrams. Pair it with the VS Code draw.io extension and watch diagrams update in real time as the AI works.
What can it do?
- Create diagrams — flowcharts, architecture diagrams, sequence diagrams from natural language
- Read & understand — parse existing
.drawiofiles and describe their structure - Edit precisely — add/remove/update individual nodes and edges
- Undo mistakes — full history with undo/redo, independent of the editor's Ctrl+Z
- Live preview — changes appear instantly in the VS Code draw.io editor, no reload
Install
Click one of the badges above, or add this to .vscode/mcp.json in your project:
{
"servers": {
"drawio": {
"command": "npx",
"args": ["-y", "github:abossard/drawio-mcp"]
}
}
}
Tip: Commit this file to your repo — every contributor gets diagram support automatically.
<details> <summary><strong>Claude Desktop</strong></summary>
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"drawio": {
"command": "npx",
"args": ["-y", "github:abossard/drawio-mcp"]
}
}
}
</details>
<details> <summary><strong>VS Code user settings (all projects)</strong></summary>
Add to your User settings.json (<kbd>Cmd</kbd>+<kbd>Shift</kbd>+<kbd>P</kbd> → Preferences: Open User Settings (JSON)):
{
"mcp": {
"servers": {
"drawio": {
"command": "npx",
"args": ["-y", "github:abossard/drawio-mcp"]
}
}
}
}
</details>
Usage
Just ask your AI assistant:
"Create a flowchart for a user login process in
docs/login-flow.drawio"
"Read
architecture.drawioand describe the components"
"Add a Redis cache node between the API and Database, then highlight it"
"Undo the last change"
The server also ships with prompt templates — try asking to "use the architecture prompt" or "use the flowchart prompt".
Tools
Diagram manipulation
| Tool | Description |
|---|---|
create_diagram |
Create a new .drawio file |
read_diagram |
Parse and return all pages, nodes, and edges |
add_node |
Add a shape with label, position, size, and style |
add_edge |
Connect two nodes with optional label and style |
update_element |
Modify label, style, position, or size |
remove_element |
Delete a node or edge by ID |
add_page |
Add a new page/tab |
get_diagram_styles |
List all predefined styles |
History
| Tool | Description |
|---|---|
undo_last_operation |
Restore file to previous state |
redo_last_operation |
Re-apply last undone change |
get_change_history |
List recent operations |
Visual feedback
| Tool | Description |
|---|---|
highlight_element |
Flash elements with a colored highlight |
show_ai_cursor |
Show AI cursor position † |
show_ai_selection |
Highlight AI-selected cells † |
show_status |
Status bar message † |
show_spinner |
Loading spinner † |
† Requires the companion extension.
Predefined styles
Use the shape parameter in add_node or edgeStyle in add_edge:
| Category | Styles |
|---|---|
| Shapes | rectangle roundedRectangle ellipse diamond parallelogram hexagon triangle cylinder cloud document process star callout |
| Flowchart | start end decision processStep inputOutput |
| Architecture | server database firewall user container |
| UML | actor component package interface |
| Colors | blue green red yellow purple orange gray |
| Edges | straight orthogonal curved entityRelation dashed dotted bidirectional noArrow |
How it works
AI Assistant ──(MCP)──> drawio-mcp server ──(writes XML)──> .drawio file
│
VS Code draw.io extension
detects change & live-merges
│
draw.io webview
(instant update ✨)
- ✅ No page reload — changes merge seamlessly
- ✅ Undo works — each write becomes one Ctrl+Z step
- ✅ User selection preserved — the merge doesn't disrupt anything
Companion Extension (Optional)
The companion VS Code extension adds real-time AI presence overlays to the draw.io editor. Without it the core MCP tools (create, read, edit, undo) work fine — the extension adds the visual extras.
What it adds
| Feature | Description |
|---|---|
| 🤖 AI cursor | Shows where the AI is "looking" in the diagram |
| 🔲 Cell selection | Highlights which cells the AI is editing |
| ✨ Flash highlight | Temporary colored flash on elements |
| 💬 Status messages | Shows AI status in the editor (e.g. "Adding database…") |
| ⏳ Spinner | Loading indicator during long operations |
| 📐 Auto-layout | Applies layout algorithms (hierarchical, organic, tree, circle, radial) |
Prerequisites
- draw.io VS Code extension — install from the marketplace or run:
code --install-extension hediet.vscode-drawio # or for Insiders: code-insiders --install-extension hediet.vscode-drawio
Install the Bridge Extension
# From the drawio-mcp repo root:
cd vscode-drawio-mcp-bridge
npm install
npm run build
npx @vscode/vsce package --allow-missing-repository
This produces a drawio-mcp-bridge-0.1.0.vsix file. Install it:
code --install-extension ./drawio-mcp-bridge-0.1.0.vsix
# or for Insiders:
code-insiders --install-extension ./drawio-mcp-bridge-0.1.0.vsix
Alternatively, in VS Code: <kbd>F1</kbd> → "Extensions: Install from VSIX…" → select drawio-mcp-bridge-0.1.0.vsix.
Verify
- Reload VS Code (Cmd+Shift+P → "Developer: Reload Window")
- Open any
.drawiofile - A small activity log panel appears in the top-left of the draw.io editor showing connection status
- Use the MCP tool
check_connectionto confirm the bridge is connected
The extension auto-connects to the MCP server's WebSocket sidecar when you open a .drawio file. No extra configuration needed.
Settings
| Setting | Default | Description |
|---|---|---|
drawioMcpBridge.websocketPort |
9219 |
WebSocket port (must match DRAWIO_MCP_SIDECAR_PORT) |
drawioMcpBridge.autoConnect |
true |
Connect automatically when a .drawio file opens |
drawioMcpBridge.aiColor |
#D13913 |
Color for AI cursor and selection overlays |
drawioMcpBridge.aiLabel |
🤖 AI |
Label shown next to the AI cursor |
Commands
- Draw.io MCP: Connect to Server — manually connect to the sidecar
- Draw.io MCP: Disconnect from Server — disconnect
How it works
┌─────────────────────────────────────────────────┐
│ VS Code │
│ │
│ draw.io extension drawio-mcp-bridge │
│ ┌──────────────┐ ┌─────────────────┐ │
│ │ draw.io │◄─plugin─┤ Injects JS into │ │
│ │ webview │ │ draw.io iframe │ │
│ └──────────────┘ └────────┬────────┘ │
│ │ WebSocket │
│ ┌──────────▼────────┐ │
│ │ drawio-mcp server │ │
│ │ sidecar (:9219) │ │
│ └───────────────────┘ │
└─────────────────────────────────────────────────┘
The extension injects a plugin into the draw.io iframe that opens its own WebSocket to the MCP sidecar (ws://127.0.0.1:9219/drawio). AI actions (cursor moves, highlights, status) flow through this channel. User interactions (mouse position, cell selection) are sent back to the MCP server so the AI knows what the user is looking at.
Development
git clone https://github.com/abossard/drawio-mcp.git
cd drawio-mcp
npm install
npm test
Point VS Code at your local build — add to .vscode/mcp.json in any project:
{
"servers": {
"drawio-dev": {
"command": "node",
"args": ["/absolute/path/to/drawio-mcp/build/index.js"]
}
}
}
Run npm run dev for watch mode — VS Code auto-restarts the server on each rebuild.
| Command | Description |
|---|---|
npm run build |
Compile TypeScript |
npm run dev |
Watch mode (rebuild on save) |
npm test |
Run tests |
npm run test:watch |
Watch mode for tests |
Environment Variables
| Variable | Default | Description |
|---|---|---|
DRAWIO_MCP_SIDECAR_PORT |
9219 |
WebSocket port for companion extension |
DRAWIO_MCP_NO_SIDECAR |
1 |
Set to 1 to disable the WebSocket sidecar |
License
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。
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。
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。
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。
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。
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。
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。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。