icebird1998/drawio-scientific-illustrator

drawio-scientific-illustrator is a plugin for Codex, an AI coding assistant, that lets the AI draw scientific figures live inside the actual draw.io desktop application. Instead of generating a finished file and opening it afterward, the AI controls draw.io's own graph tools directly, so a person can watch shapes, labels, arrows, and layout appear on screen step by step as the drawing is built. The project is important to note as legacy: the README states clearly that development has moved to a new repository called scientific-illustrator, and this older version no longer receives feature updates or compatibility fixes, though it remains publicly available for existing users. New and existing users are pointed toward the newer project instead. Under the hood, this repository combines two pieces. First, it includes MCP servers, small background programs that expose callable tools, one that connects to the visible draw.io editor and edits its graph in real time, and another that validates saved drawio files and exports them as PNG, SVG, PDF, or JPG images. Second, it includes a Codex skill, instructions that teach the AI agent how to look at a reference image, break it down into simple editable shapes, draw them with pacing so the process is visible, review sections visually, refine what it drew, and only save the file once the drawing already exists on screen. To use it, someone needs the Codex desktop app or CLI with plugin support, draw.io desktop installed locally, Git, and Node.js version 22 or newer. Installation can happen through a one line script for Windows, macOS, or Linux, or by cloning the repository and registering it manually as a Codex plugin marketplace. Once installed, a user attaches a reference image such as a PNG or PDF page, mentions the plugin by name, and describes the pacing and output formats they want, and the AI redraws the figure inside draw.io using a fixed sequence of tools: launching the editor, checking its status, adding shapes and edges, taking screenshots to inspect progress, correcting details, and finally saving and exporting the result. The project is released under the MIT license.