icebird1998/drawio-scientific-illustrator

This open-source project exposes the visible draw.io canvas to MCP control, letting a Codex-side agent issue step-by-step drawing instructions that the user can watch execute in real time. The scientific illustration framing is specific: this is not a general diagram-from-prompt tool but a system designed for figures that need to be constructed with precision — pathway diagrams, architecture schematics, that kind of thing. The live observation of each step is the practical value; it makes the agent's decisions auditable and correctable before the diagram is finished, which matters when the output goes into a paper or a presentation. Reservation: the audience here is narrow — you need to be a draw.io user, working in Codex, and producing the kind of structured scientific content that benefits from agent-assisted construction. If any of those three conditions does not hold, there is not much here. -> Best for: ML researcher or technical PM who regularly produces scientific figures and already uses draw.io