Bridging AI and Design Files With MCP
Penpot is publicly experimenting with Model Context Protocol (MCP) servers, a move that could let AI assistants directly understand and act on design files. The core idea: an AI like Claude or Gemini can’t natively read a Penpot document, but an MCP server acts as a translator. It takes the AI’s intent, picks the right operation, and converts it into a structured call against Penpot’s extensive API.
The reason this works so well for Penpot is its design-expressed-as-code foundation. Every element in a file is programmatically addressable, which gives MCP servers granular control for creating, editing, and analyzing designs. Unlike vague “describe → generate” workflows, these operations are contextual and precise. Since Penpot is open-source, the integration surface is wide open, and the MCP approach keeps third-party LLMs from ever touching your data directly. The server is the secure bridge.
How the Servers Are Built
The implementation is still early-stage but already fairly complete. Penpot’s MCP server:
- Complies with current MCP standards.
- Integrates with the Penpot API for real-time design data.
- Includes a Python SDK, REST API, plugin system, and CLI tools.
- Works with any MCP-enabled AI assistant (Claude in VS Code, Cursor, Desktop, etc.).
- Supports sharing design context, including component structures, with AI models.
- Enables natural-language communication for design tasks.
What the Demos Show
Some of the most convincing evidence is in video demos shared by Penpot and community members. They show workflows that go well beyond simple export or text generation.
Design as Transmutable Data
Because Penpot files are code under the hood, design assets can be converted in multiple directions: design to code, code to design, design to documentation, and documentation back into design system elements. Penpot designer Juan de la Cruz García demonstrated this by turning simple components into docs, design system tokens, new components, and even a complete Storybook project, all driven by prior design decisions rather than vague prompts.
Complex, Multi-Step Operations
Another demo from Dominik Jain of Oraios AI shows a more involved session. He created a Node.js web app from a design, updated frontend styles, stored names and identifiers in memory for consistent translation, validated the result, added comments next to selected shapes, and replaced a scribble with an adapted component. The whole session ran through Claude Desktop, showing how robust the interaction can be.
Notably, Penpot MCP is LLM-agnostic. One community member, IvanTheGeek, set it up with the JetBrains Rider IDE and Junie AI, which points to how flexible the setup can be depending on your stack.
Practical Use Cases Already Emerging
Beyond the headline demos, Penpot and its community have outlined a broader set of tasks they’re targeting:
- Translating boards to production-ready semantic HTML and modular CSS, including the use of existing design tokens.
- Generating interactive web prototypes without altering existing HTML.
- Converting scribbles into components based on the current design system.
- Building design system documentation directly from a file.
- Advanced export operations (e.g., exporting only the icons actually used).
- Natural-language search for design elements.
- Pulling data from external APIs via conversation.
- Connecting Penpot to other external tools.
- Memorizing and repeating repetitive tasks.
- Visual regression testing and design consistency checks.
- Accessibility/Usability analysis and feedback.
- Design system and brand/content guideline compliance checks.
- Design analytics to monitor component adoption.
- Keeping documentation synced with design automatically.
- File organization like tagging and categorization.
The long-term pitch is that these servers could enable a nearly infinite number of workflows, combining the efficiency of your chosen LLM client with the security of never exposing your data directly to it.
An Open Experiment Looking for Testers
Penpot MCP is not in beta yet, but it’s an active experiment, and the company is inviting users to help steer development. Penpot’s CEO Pablo Ruiz-Múzquiz has pointed to several demo videos (03, 04, 06, 08, and 12) as personal favorites for those wanting a quick overview.
The call is for collective input. Penpot argues that design tools of the next generation need to be built collaboratively, especially around AI, and they want to hear about use cases current tools can’t handle well, quickly, or at all.
If you want to test the waters, Penpot has published setup instructions for its MCP server, and you can join the community discussion to share feedback or ideas.
Note: Penpot is actively seeking beta testers for the MCP server. Email [email protected] with the subject line “MCP beta test volunteer” to get involved.
