Control Grasshopper with Claude and MCP

A useful tool describes a meaningful operation. It tells the agent what it can do, which inputs are required and how to interpret the result. This is easier to reason about than granting access to a large list of anonymous controls.

Choose the interface around the design task. Some parameters should be adjustable, others fixed, and some outputs available only for inspection. Making those choices explicitly helps keep the agent's actions within the model's intended scope.

An instruction to change a value is only half an interaction. The agent also needs to know what happened afterward. Expose outputs that describe whether the operation succeeded and how the design changed.

The right feedback depends on the task. A metric may support comparison, while geometry provides context that the metric omits. Keep human review in the loop when the design question involves qualities that are not represented in the exposed outputs.

Begin with a known model state and make one change at a time. Inspect whether the tool description, accepted values and returned information match reality. Once the round trip is dependable, more complex sequences become easier to assess. The connection is valuable when it makes a model easier to operate without hiding how it works.