drawio-mcp-server
Enables creating and editing draw.io diagrams from structured descriptions, rendering them as interactive inline SVGs, and providing access to the full draw.io editor for further modification and export.
README
drawio-mcp-server
An MCP (Model Context Protocol) server for draw.io / diagrams.net with an interactive, editable diagram UI.
It exposes tools to build a diagram from a structured description (nodes + edges) and render it on an embedded, editable draw.io canvas right inside the client — where the user can keep editing and export to PNG, SVG, or XML.
Modeled on googlemaps-mcp-server: same MCP-App pattern (data tools return content, a UI tool returns an embedded interactive HTML resource), but no authentication — draw.io is fully client-side, so there are no API keys and no OAuth.
Prerequisites
- Node.js 18+
- pnpm
Setup
- Install:
pnpm install
- Build and run:
pnpm dev
The server starts on port 3000 and exposes the MCP endpoint at /mcp.
Environment Variables
| Variable | Required | Description |
|---|---|---|
PORT |
No | Port to listen on (default: 3000) |
Endpoints
| Endpoint | Description |
|---|---|
/mcp |
MCP endpoint (GET, POST, DELETE) — no auth required |
/health |
Health check |
MCP Tools
create_diagram
Build a draw.io diagram from a structured description and return valid mxGraph XML.
Parameters:
nodes(required) — array of{ id, label?, shape?, x?, y?, width?, height?, fillColor?, strokeColor? }.shapeis one ofrectangle,rounded,ellipse,diamond,process,terminator,cylinder,cloud,hexagon,parallelogram.edges— array of{ source, target, label?, dashed? }referencing node ids.direction—vertical(default) orhorizontal, used by the auto-layout when positions are omitted.
Positions are auto-laid-out (layered, centered) when x/y are omitted — just describe what connects to what.
render_diagram
Display the diagram inline. Renders ready-made mxGraph xml (typically the output of create_diagram); building from a description lives in create_diagram, so this tool only renders.
Parameters: xml (the diagram to render — omit for an empty placeholder), title (optional).
The diagram renders as interactive inline SVG (zoom / pan / fullscreen lightbox) via the draw.io viewer script, plus an "Edit in draw.io" button that opens the full web editor in a new tab with the diagram preloaded (where you can edit and export to PNG/SVG/XML).
Why a script, not an iframe: MCP Apps render inside a sandboxed iframe that allows in-document scripts but blocks nested iframes (CSP
frame-src 'none'). So the UI loads the draw.io viewer (viewer.diagrams.net/js/viewer-static.min.js) as a<script>and draws inline SVG — the same mechanism the google-maps MCP app uses to render its map. Embeddingembed.diagrams.netas a nested iframe renders blank.
Resources
ui://drawio/editor
The diagram UI rendered by render_diagram, served as an MCP App resource. It loads the draw.io viewer as an in-document script and draws the diagram as inline SVG.
Client Configuration
{
"mcpServers": {
"drawio": {
"type": "streamable-http",
"url": "http://localhost:3000/mcp"
}
}
}
For a deployed server, replace the URL with your public endpoint, e.g. https://your-app.up.railway.app/mcp.
Deployment
The server is stateless and keyless — deploy the built dist/ and run pnpm start behind any HTTPS host (Railway, Fly, etc.). The only configuration is the optional PORT.
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。