Draw.io Scientific Illustrator: A Legacy Codex Plugin for Live Canvas Drawing

What Draw.io Scientific Illustrator does and its current status

Draw.io Scientific Illustrator is a Codex plugin that enables an AI agent to draw scientific figures inside the visible draw.io desktop editor, with shapes, labels, arrows, and styling appearing step by step as the agent works. The README describes the mechanism as calling draw.io's own graph API through a localhost-only MCP server, rather than automating the operating-system mouse or keyboard or generating XML first and opening it afterward.

The repository's README opens with a migration notice: all future development, releases, and fixes have moved to the scientific-illustrator project at github.com/icebird1998/scientific-illustrator. This repository remains publicly accessible and is not being archived, but it no longer receives feature or compatibility updates. The README recommends that both new and existing users download the new project.

For users who have already installed this version and do not wish to migrate, the plugin continues to function as documented. Windows was the tested platform for this version; the successor project adds cross-platform support.

The repository had one release, v1.0.0, tagged on 2026-07-12, approximately three weeks before the last push. The migration notice and legacy designation were added after that release, so v1.0.0 represents the final feature-complete state of this version. The CHANGELOG.md tracks changes prior to that point.

What the repository contains: two MCP servers and a Codex skill

The repository packages four components as a Codex plugin. The drawio-live MCP server launches or connects to the visible draw.io desktop editor and edits its active graph model in real time. The drawio-file-utils MCP server validates saved .drawio documents and exports PNG, SVG, PDF, or JPG files.

The Codex skill teaches the agent how to inspect a reference image, decompose it into editable draw.io primitives, draw with pacing, visually review sections, refine the live graph, and save only after the visible drawing exists. A repository-local Codex marketplace entry allows installing the complete plugin as one unit.

The README describes this as simultaneously an MCP implementation and a Codex plugin: MCP provides the callable tools, and the Codex plugin is the installable package containing the MCP servers, skill, and presentation metadata.

The package.json names the package drawio-scientific-illustrator-marketplace at version 1.0.0, with node >=22 as the engine requirement. The scripts field provides a check target that validates the three core MCP server scripts and a smoke-test target for basic validation.

Requirements and installation paths

The plugin requires: the Codex desktop app or Codex CLI with plugin support, draw.io desktop installed locally, Git, and Node.js 22 or newer for running the MCP servers outside the bundled Codex runtime. The plugin auto-detects common draw.io installation paths on Windows, macOS, and Linux; a DRAWIO_PATH environment variable overrides the auto-detection.

The README documents two one-command installers. For macOS and Linux:

curl -fsSL https://raw.githubusercontent.com/icebird1998/drawio-scientific-illustrator/main/install.sh | bash

For Windows, after reviewing install.ps1:

$p="$env:TEMP\drawio-scientific-install.ps1"; Invoke-WebRequest https://raw.githubusercontent.com/icebird1998/drawio-scientific-illustrator/main/install.ps1 -OutFile $p; powershell -ExecutionPolicy Bypass -File $p

For an auditable manual install on any platform:

git clone https://github.com/icebird1998/drawio-scientific-illustrator.git
cd drawio-scientific-illustrator
codex plugin marketplace add "$(pwd)"
codex plugin add drawio-scientific-illustrator@drawio-scientific-tools

After installation, Codex must be restarted and a new task started so the skill and MCP tools load. The README notes that the one-command installers do not ask the migration question and install the legacy version directly; they should be used only after explicitly choosing to stay on the legacy plugin.

Live tool workflow and the agent's drawing sequence

The agent follows a documented tool sequence when drawing a scientific figure. The README describes the standard order as:

1. drawio_live_launch to connect to or start the visible editor 2. drawio_live_status to confirm the graph is ready 3. drawio_live_add_shape, drawio_live_add_edge, or drawio_live_draw_sequence to build content 4. drawio_live_screenshot to inspect the renderer after each section 5. drawio_live_inspect and drawio_live_update_cell to correct labels, styles, and positions 6. drawio_live_fit to keep the evolving figure visible 7. drawio_live_save_snapshot to serialize the visible graph to a .drawio file 8. drawio_validate and export tools to verify and produce deliverables

The recommended starting prompt asks the agent to use a 100 ms delay between steps so the drawing is visually traceable:

Use Draw.io Scientific Illustrator. Launch the live draw.io canvas and recreate this
reference scientific figure step by step with a 100 ms delay. Control only draw.io's
own graph API; do not use OS mouse/keyboard automation and do not generate XML first.
Keep all labels, arrows, panels, and legends editable. Visually inspect and refine each
section, then save the final .drawio file and export a 2000 px PNG preview.

The README recommends using GPT-5.6 Sol with Max reasoning effort for complex scientific redraws. It notes that the six-level reasoning selector must be enabled in Codex settings first, as the default five-level selector does not expose the Max option.

Limitations of the legacy version

The primary limitation is maintenance: this repository no longer receives feature or compatibility updates. draw.io desktop receives regular updates; changes to its graph API may break the MCP server without a fix being available in this repository. The successor project scientific-illustrator is where such fixes would appear.

The README states that Windows was the tested platform for this version. While the installer scripts exist for macOS and Linux, cross-platform reliability is not guaranteed and the successor project is the documented path for non-Windows use.

Because the MCP server runs locally and communicates with the draw.io desktop app through localhost, the plugin requires draw.io desktop to be installed and running. The web-based draw.io editor at app.diagrams.net is not supported; there is no hosted mode where the MCP server connects to a remote draw.io instance.

A PRIVACY.md file is included in the repository and covers what data the MCP server processes locally; it should be read before deploying the plugin in an environment with sensitive research material. The .agents/ directory contains agent configuration for the Codex plugin, and the plugins/ directory holds the plugin manifest and MCP server scripts that are registered with Codex during installation.

For users who want a maintained alternative outside the draw.io ecosystem, Inkscape with an LLM-assisted SVG generation workflow offers a different approach: SVG is text-based and can be generated or edited programmatically, but Inkscape does not provide a live step-by-step visual workflow comparable to this plugin's canvas interaction model. The draw.io Scientific Illustrator's specific contribution is the step-by-step live visibility as the figure is built, not just the final output.

Editorial conclusion

Researchers and engineers who need an AI agent to draw scientific figures live inside draw.io should install scientific-illustrator, the project's successor at github.com/icebird1998/scientific-illustrator, which receives current updates and cross-platform support. This repository remains available for users who installed it before the migration and do not want to switch, but the README explicitly states it no longer receives feature or compatibility updates. The install path for new users still works, and the two MCP servers, drawio-live and drawio-file-utils, function as documented on the last push date of 2026-07-30.

Frequently asked questions

Official sources

1. icebird1998/drawio-scientific-illustrator on GitHub
2. Issues
3. License: MIT
4. README
5. Releases