Until recently, AI coding assistants could only generate static CSS snippets or generic code templates when asked to create user interfaces. The Model Context Protocol (MCP) changes this by giving AI agents direct control over visual design canvases and state machines.
Why static code generation falls short for rich UI
When an AI generates 400 lines of raw React JSX containing custom physics math and bezier easing arrays, the result is difficult to preview, brittle to maintain, and impossible for visual designers to edit without coding.
How the Model Context Protocol (MCP) works
The Model Context Protocol is an open standard developed by Anthropic that allows local AI tools (such as Claude Code CLI, Claude Desktop, Cursor, and Windsurf) to securely communicate with local applications over standard input/output (stdio).
Instead of hallucinating code, the AI agent calls specific tools provided by an MCP server to read file schemas, inspect active viewports, and manipulate structured project files.
The ExodeUI MCP server workflow
ExodeUI, which we make, publishes an official MCP connector on npm as `claude-exode-mcp` (exposing 149 tools). When connected:
1. **Prompt to Artboard:** You prompt Claude Code: *"Create a bottom navigation bar with 4 tabs and wire an active tab state machine."*
2. **Tool Execution:** Claude executes tools like `scaffold_component`, `add_input`, and `add_object_interaction` to author the structure.
3. **Live Sync:** The generated layers appear in real time inside your open ExodeUI editor on your machine.
4. **Runtime Code:** The output is a standard .exode JSON file that your web or mobile app plays through the `exodeui` runtime.
Visual refinement meets agentic speed
This workflow gives teams the best of both worlds: the speed of conversational AI generation paired with the precision of a visual editor where designers can fine-tune spring damping, colors, and layout by hand.