Diagrams via chat, because drag-and-drop is slow

What it does

Next AI Draw.io is a Next.js application that wraps draw.io inside an AI chat layer. You describe what you need—say, a RAG pipeline or an AWS serverless auth flow—and the app streams an LLM’s output into draw.io XML that renders live in the browser. It can also ingest images, PDFs, and text files to replicate or summarize them as diagrams, and it keeps version history so you can view and restore previous versions before the AI editing.

The interesting bit

The project treats draw.io XML as a strict intermediate language for LLMs, which is harder than it sounds; the README explicitly warns that weak models struggle with the formatting constraints. It also ships an MCP server, so Claude, Cursor, or VS Code can spawn and edit diagrams in your browser without you touching the UI.

Key highlights

• Supports 14+ LLM providers, from OpenAI and Anthropic to Ollama and ByteDance Doubao, with most allowing custom endpoints.
• Can generate animated connectors and cloud architecture diagrams (AWS, GCP, Azure); the README notes claude models are particularly good at cloud logos.
• Offers an MCP server for agent integration, turning the app into a real-time whiteboard for AI coding assistants.
• Diagram history tracks all changes, letting you roll back to earlier versions.
• Ships as a web app, desktop app (Windows/macOS/Linux), or deployable to EdgeOne Pages, Vercel, and Cloudflare Workers.

Caveats

• The README notes that generating valid draw.io XML requires strong model capabilities, so budget or weak models may not fare well.

Verdict

Worth a look if you spend too much time clicking through cloud architecture stencils or want your AI agent to sketch while it codes. Skip it if you need guaranteed pixel-perfect diagram fidelity without human cleanup.