Dimensions, colour, shape and description are written into each PNG as JSON and read back through read_image_metadata, which is what turns a folder of coloured rectangles into something an agent can reason about a week later. Batching covers spritesheets as well as individual files, with margin and spacing as parameters, so packing is not a separate step with a separate tool. The reader is scoped to what this server produced — it reads the metadata this generator embedded, not arbitrary PNGs from elsewhere.
A mock asset generator for game prototypes: single PNGs or a batch, each carrying its own JSON metadata inside the file.
- generate_mock_asset, which draws one PNG from a shape, primary and secondary colours with a gradient angle, a pattern with its own colour, opacity and scale, a stroke, and text with a size, colour, position and rotation, at a given width and height.
- generate_mock_asset_batch, which produces many at once either as individual files or as a spritesheet with its own margin and spacing.
- JSON metadata embedded in every generated PNG — dimensions, colour, shape and description — read back through read_image_metadata.
Nothing to supply — no account, no key. A directory for the PNGs to be written into, given per call as directory or sheetDirectory.
One command — npx -y 2d-assets-mcp