VectorCraft: Create a Scalable Graphic

A water-refill sign ought to be a simple job. Draw a droplet, add a clear label, and make something you can use on a website or enlarge for a noticeboard.

It is also a useful test of a vector editor. The goal is not just a nice-looking picture. It is a graphic whose shapes remain editable and whose edges do not depend on the resolution of the first export.

For this Spotlight, I built a fresh refill-station graphic in VectorCraft, ArtCraft’s free, open-source vector illustration app. Then I checked two ways of working with it: desktop editing and an agent-facing connection. They are related capabilities, but one successful export does not prove both.

Why this job wants vectors

A PNG stores a grid of pixels. Make that same PNG much larger and those pixels eventually become visible. A vector file describes shapes and curves, so software can draw them again at the size you need.

That is useful for logos, icons, labels and simple signs. It does not mean every vector design works at every size. Small lettering can still become unreadable, and a crowded layout remains crowded. Vectors solve the resolution problem, not the design problem.

Last week’s PhotoCraft Spotlight explored an image editor. VectorCraft makes a natural next stop: instead of starting with a photograph, this task starts with a few shapes and some type.

A finished graphic, from a fresh document

The example uses a 640-by-640 artboard: a warm off-white border, a rounded dark-teal panel, a mint droplet and two lines of text. It is an original demonstration graphic, not an imported template or an existing personal asset.

I created the master with the installed VectorCraft 0.7.0 command-line tool. Its document inspection showed two editable type objects, two paths and three rectangle-based shapes, inside one layer. The droplet is a curved path, not a pasted bitmap.

I saved and reopened the native .vectorcraft file, then exported PNGs at 160, 640 and 1,920 pixels square. Their dimensions checked out. At the smallest size, the droplet and headline still read clearly, although the supporting line is small enough that I would simplify it for a tiny website icon.

The SVG export was a more important check. It contained vector geometry with no embedded images. I deliberately converted the lettering to outlines for that export, making the letter shapes independent of an installed font. The tradeoff is that those letters are no longer editable as ordinary text in the SVG; the native master retains the editable type. Reopening the SVG and rendering it at 1,920 pixels also succeeded.

Checking the desktop separately

A command-built file is not evidence of a good mouse-and-keyboard workflow. I opened the master in the desktop app and checked a smaller, specific sequence: select and drag the droplet, undo the move, edit the supporting line with the Type tool, save a separate copy, and reopen it.

Those checks used VectorCraft’s UI control channel to send pointer and keyboard events. They exercised the desktop tools, rather than replacing the text directly through a document command. They were automated UI checks—not a claim that I drew the whole sign by hand or assessed a long illustration session.

The moved droplet returned to its original position after undo. The edited line survived saving and reopening the separate copy. That establishes a basic edit-and-return workflow for this document, not a verdict on every tool in the application.

If you prefer to work without an agent, the useful starting points are familiar: create an artboard, draw shapes, set their fills, add type, and keep an editable master before exporting. The Pen and Direct Selection tools are where a simple shape becomes a more individual mark.

Give an agent a bounded variation

VectorCraft also supplies an MCP server—a connection that lets compatible AI agents call the editor’s tools. Its documentation distinguishes a connection to the running desktop app from a headless mode that works without a window.

I tested the headless MCP server directly: initialize a connection, discover its tools, open the saved master, replace the supporting line with “BRING A BOTTLE,” center that line on the artboard, and save and export a separate variation. Reopening the variation confirmed that its text remained editable. The original master was left unchanged.

This was a test of actual MCP tool calls, not just a feature listed in a README. It was not a test of a one-sentence prompt autonomously producing the right design, nor of MCP controlling the live desktop. An agent still needs a clear brief and someone checking its output. The editor is free; whichever AI service you connect may have its own costs.

Where I would start—and where I would stop

This narrow workflow worked: make a small original graphic, retain editable shapes and text, export scalable geometry, edit it in the desktop, and produce a controlled variation through agent tools.

It does not establish Illustrator parity, reliable import of a complicated client file, or production print readiness. This example is RGB artwork. I did not test a commercial printer, color separation, font licensing for redistribution, or an entire brand package.

Version matters, too. These hands-on checks used the installed 0.7.0 app and CLI, with the matching source and documentation. At the time of writing, the official download page offered 0.9.0. This is evidence about the tested build, not a review of every change in the newer release.

For a first experiment, choose a modest job with a real destination: a refill sign, a club badge or a simple service icon. Save the native file, export an SVG, and inspect the result at both its smallest and largest intended size. Then try one bounded change through an agent. You will learn more from finishing that loop than from counting buttons.