Drawio Mcp Server 🔌 MCP server Free

What it does

Draw.io MCP Server is a Model Context Protocol server that lets AI coding agents create, inspect, and edit diagrams directly inside draw.io (diagrams.net). Instead of describing a diagram in words, an agent can call the server to add shapes, wire up connectors, organize pages and layers, and read back the current state of a diagram — useful for auto-generating architecture diagrams, flowcharts, or system documentation as part of a coding session.

What is inside / Tools it gives your agent

The server exposes MCP tools for document discovery across multiple open tabs, shape and page inspection (reading existing elements, layers and properties), diagram editing (adding, updating and deleting shapes and connectors), page/layer management (create, copy, rename), vendor stencil lookup (AWS, GCP, Azure, Cisco shape libraries are auto-discovered), import/export to XML, SVG and PNG, and Mermaid diagram conversion.

Works with

Listed as "*" — any MCP-capable client. The README specifically documents Claude Desktop, Claude Code, Zed, Codex and OpenCode, connecting either through a built-in local editor (http://localhost:3000) or a browser extension that attaches to a draw.io tab.

How to install or connect

For Claude Code, install directly as a plugin: /plugin marketplace add lgazo/drawio-mcp-server. For other MCP clients, run npx drawio-mcp-server install <host>, or add npx -y drawio-mcp-server --editor to your client's MCP config manually. Node.js v22+ is required.

Maintenance and safety

MIT licensed and actively maintained (1,400+ stars, recent commits). Runs locally; supports optional TLS with self-signed certificates, and serializes edits per-document over WebSocket to avoid collisions when multiple agents touch the same diagram. Desktop-app integration is still experimental due to upstream content-security-policy restrictions.

Who should use it

Developers and architects who want their coding agent to produce or update real, editable draw.io diagrams — architecture diagrams, ERDs, flowcharts — instead of static images or text descriptions.