AI agent integration
Isometic MCP for AI Agents
Isometic Studio exposes native editor operations through the Model Context Protocol (MCP), allowing compatible AI clients to inspect a live scene, create structured geometry, arrange objects and render the current result for visual feedback.
What MCP changes
Without an editor integration, an AI model can only describe an illustration or generate a flattened image. With Isometic's MCP tools, the client works with the editor's native objects instead. Shapes remain selectable and editable, relationships can be measured, and the result can still be exported through the normal Isometic workflow.
Supported workflow
Inspect the scene
Read structured shape data, bounds, features, groups and the current document revision.
Render the result
Return the editor's native PNG so a vision-capable client can inspect the actual illustration.
Create geometry
Create primitives, custom paths, advanced shapes, lines and editable text.
Arrange objects
Align, distribute, stack, snap, attach, transform and reorder scene objects.
Use assets
List and insert reusable saved assets, then combine them with generated geometry.
Edit safely
Use revision-aware operations, batches, dry runs and the editor's normal undo/redo history.
Typical agent loop
- Inspect the current document and available editor capabilities.
- Create or insert the main objects needed for the requested illustration.
- Arrange the objects with explicit alignment, stacking, snapping or attachment operations.
- Render the scene and inspect the real image rather than assuming the geometry looks correct.
- Refine the scene and verify important spatial relationships before finishing.
Why structured editing matters
The MCP integration operates on native Isometic objects rather than pixels. That means an agent can build a scene that a person can continue editing afterward: moving a layer, changing text, recoloring a shape, replacing an asset or exporting the composition as SVG or PNG.
Important limitation
MCP provides the editor interface; it does not guarantee design quality. A capable client still needs to interpret the prompt, inspect rendered output and decide whether the composition is visually correct. Some advanced geometry and attachment cases also have narrower support than basic primitives and planar faces.