The song that gives this release its title is an not-so-subtle nod to the only (but major) feature of this release: the Penpot MCP Server.
Penpot MCP moves from an early, more technical setup to a much simpler in-app experience with remote configuration, requiring just a few clicks to get started. This is where it begins to feel effortless to connect your AI client and turn prompts into real actions on real design data.
With this release we are opening the door to truly multi-directional workflows between design and code, while staying faithful to Penpot values: openness, freedom of choice, and respect for your data.
Penpot MCP Server: AI connected to real design context
Penpot MCP Server is the bridge between your AI client and your Penpot file. You describe what you need in natural language, your agent picks the right operation, and MCP translates that into real actions through Penpot APIs.
This is not a generic “describe and generate” flow. It is context-aware work with components, tokens, pages, layers, and structure. In short: design expressed as code, now usable through the AI assistant of your choice.
You can run MCP in two ways. Remote MCP is hosted and simpler to set up. Local MCP runs on your machine and gives extra tools (mainly access to the local file system).
Multi-directional workflows, from design to code and back
The biggest unlock is multi-directionality. You can move from design to code, and from code back to design, without losing intent or structure in the process. These are just some of the more common use cases we’re already seeing in practice:
• Generate semantic HTML/CSS from real layouts.
• Create design tokens and apply them consistently.
• Export assets in use.
• Generate component variants.
• Validate design-code consistency.
• Reorganize layers, apply naming rules, and automate repetitive design system maintenance.
• Establish a direct, bidirectional connection between your Penpot files and Storybook (or similar tools).
The MCP becomes workflow infrastructure: less manual glue work, fewer handoff gaps, and faster iterations between designers and developers.
This is one of my favorite examples. I think it’s because it reminds me of those early experiments (8 years ago already!) by the Airbnb team, where they used AI to build functional prototypes from rough sketches. Can find many more cool use cases in the Penpot MCP video playlist.
Your stack, your model, your control
With MCP, you connect Penpot to the AI client and model of your choice. Cursor, Claude, VS Code, Codex, or another MCP-compatible setup. Also, you can run it hosted, or locally. The same applies to data boundaries: Penpot provides the bridge to your design context, while your team decides how and where AI runs.
In practice, this means teams can automate design and code workflows without giving up tool freedom, deployment control, or ownership of their process.
You can find all the information you need to use it locally or remotely in the user guide. Enjoy.
