GH Agent

Parametric design workflows in Grasshopper are powerful but opaque — a canvas can hold dozens of interdependent components, and when something breaks, there is no contextual help without leaving the application entirely. GH Agent is an MCP connector that bridges Claude Desktop with Grasshopper 3D, reading the live canvas state so Claude can explain what’s happening, diagnose errors, add components, or write scripts — all from a conversation. Free for students and educators, no API configuration required.

GH Agent was built without prior programming experience. I collaborated with Claude throughout — not as a code generator, but as a technical partner: it explained how MCP worked, reasoned through software architecture options, wrote the code, and occasionally pushed back to sharpen the goal. I drove the project through design instinct — knowing what the tool should feel like, who it was for, and when a technical direction wasn’t right. What emerged was something neither of us could have built alone.

The design profession tends to frame AI in one of two ways: as a faster pencil — just another tool in the workflow — or as a threat that will eventually replace entry-level jobs. Both framings miss what’s actually at stake. Calling AI a tool domesticates it into existing categories of efficiency; treating it as a threat reduces the conversation to labor economics. Neither asks the more interesting question: what becomes possible when a designer thinks with AI as a genuine collaborator, one with capacities that are genuinely different from our own? GH Agent is an attempt to answer that question through practice — not to automate Grasshopper, but to make designers capable of going further than they could before.