Skip to main content
Lab Grimoire
TW EN
☕ Coffee
Spellbook / drawio-mcp: The Official draw.io MCP Server for AI-Native Diagramming
ai-tools

drawio-mcp: The Official draw.io MCP Server for AI-Native Diagramming

The official draw.io MCP server lets your LLM generate, preview, and export editable diagrams mid-conversation. Free, open-source, with 10,000+ shapes including AWS, Azure, GCP, and UML libraries. Best for developers and researchers who need architecture or workflow diagrams without leaving their AI chat. Hosted endpoint means zero setup; self-host if you handle sensitive data.

Best For

Developers and researchers who frequently create architecture diagrams, flowcharts, or system design documents within an AI-assisted workflow. Especially valuable if you already use draw.io and want to eliminate the copy-paste loop between your AI chat and the diagram editor.

How I Actually Use It

I have the hosted MCP endpoint (https://mcp.draw.io/mcp) configured in my Claude setup. When I need a system architecture diagram or an experiment workflow chart, I describe what I want in the conversation. The create_diagram tool renders a fully editable draw.io diagram inline. If I need a specific icon (say, an AWS Lambda symbol or a UML class box), I call search_shapes first to find the exact shape name, then feed it into the diagram XML.

The typical loop: describe the system in plain language, get a diagram back, ask for adjustments, export the final version. No tab-switching, no manual XML editing.

Where It Is Strong

  • Official and maintained: backed by the draw.io team at JGraph, shape index stays current with each draw.io release
  • Zero-install hosted option: point your MCP config to https://mcp.draw.io/mcp and you are done
  • Massive shape library: 10,000+ shapes spanning AWS, Azure, GCP, UML, network topology, and general flowchart symbols, far beyond what Mermaid or PlantUML can offer
  • Flexible input: accepts draw.io XML, CSV, and even Mermaid syntax via the Tool Server mode
  • Clear privacy story: self-host the npm package (@drawio/mcp) for fully local operation when handling confidential designs

Where It Fails

  • Hosted mode sends data to draw.io servers: if your diagrams contain proprietary architecture or patient data, you must self-host; the hosted endpoint is not suitable for sensitive content
  • Inline rendering depends on MCP Apps support: not every AI client supports MCP Apps Extension yet, so the interactive preview may not work everywhere
  • No round-trip editing from rendered output: you can generate and view, but making fine-grained tweaks (dragging nodes, adjusting bezier curves) still requires opening the diagram in the draw.io desktop app or web editor

Pricing, Difficulty, and Risk

Pricing: completely free and open-source (Apache 2.0). The hosted endpoint costs nothing. Self-hosting via npm is also free.

Difficulty: beginner-friendly. Configuration is a single JSON entry pointing to the hosted URL. Self-hosting requires npx @drawio/mcp or a Docker container, straightforward for anyone comfortable with a terminal.

Risk: the main concern is data privacy on the hosted endpoint. For non-sensitive work this is a non-issue. Stability is solid given official backing. No vendor lock-in since diagrams export as standard draw.io XML.

Verdict

If you produce diagrams as part of your AI-assisted workflow, install it. The shape library alone justifies the switch from Mermaid for anything beyond basic flowcharts. Skip it only if you never need diagrams or are unwilling to self-host for confidential content.

Source