GUIDE A
Try It: An Agent Inside Rhino
MCP · Rhino · Grasshopper · for designers and students
Length1–2 hours
NeedsRhino 7 or 8 · an agent
FilesA copy of a real project
What you will be able to do
- Connect an agent to Rhino through MCP, and say in plain words what you installed
- Direct an agent through three tasks from daily work, and check each result
This guide adapts a studio exercise for anyone who already works in Rhino, in an office, a studio or on their own. It assumes you know the program and skips the basics. You write no code: the agent writes and runs every script. You set the task and check the result, the way you would check a colleague’s work.
Any step you get stuck on, paste the error to the agent and ask it to walk you through. That is how most of this gets set up in practice.
Your agent
→
MCP
→
Rhino
→
Checked result
Part 1
Connect the agent to Rhino
01
What MCP Is Read first
A chat assistant works from your description of a model; it never sees the file. MCP, the Model Context Protocol, is an open standard that changes that. A small adapter sits beside a program and publishes what the program can do as a list of tools the agent may call. With the Rhino adapter installed, the agent reads your document, queries objects, builds geometry, and captures the viewport to see what it made.
- A standard, not one vendor’s feature. Most current agents support it: an agent that supports MCP (Claude Code, Codex, Cursor, Gemini CLI and others).
- The adapter runs on your machine. The .3dm stays where it is. Whatever the agent reads from it, though, goes to the AI provider as part of the conversation. Check your provider’s data terms before you open client or confidential work.
- You stay responsible for the output. The agent edits your real document. Work on a copy and check what changed.
MCP was published in late 2024. Some adapters are excellent and many are prototypes. The Rhino one is among the better ones.
02
Install the Bridge RH SETUP
Open your agent in any folder and ask:
Research the available Rhino MCP servers, install the best-maintained one for me, and tell me in plain language what you put on my computer.
The agent searches, most likely picks the community server rhinomcp, installs what it needs and writes its configuration. Approve the steps it asks about.
One part it hands back to you: the Rhino plug-in installs from inside Rhino. Tools → Package Manager, search rhinomcp, Install, restart Rhino.
Read its plain-language answer before you move on, and ask follow-up questions until it makes sense. Knowing what landed on your machine is what lets you install the next adapter, for QGIS or anything else, on your own.
03
Open the Port, Confirm the Link RH
Type mcpstart in Rhino’s command line every session. It opens the bridge. Restart Rhino and you type it again. Keep one agent connected at a time.
Then ask the agent:
What’s in my Rhino document?
A correct answer lists your layers and object counts. If the agent says it has no Rhino tools, restart the agent. If it says the connection failed, run mcpstart again.
Part 2
Three tasks from daily work
T1
Slope Bands on Your Own Terrain RH
Open a copy of a project with a terrain surface or mesh, then ask with your own thresholds:
Read the surface on my Existing Grade layer. Color it by slope in four bands: under 2%, 2–5%, 5–8.33%, over 8.33%. Bake the result as a mesh on a new layer called Slope Analysis, add a legend, then capture the top viewport and check that the steep zones read correctly.
The bands above are one common set: minimum for drainage, the accessible-route limit, the ramp limit. Use the ones your project needs and be ready to say why.
- Name layers exactly. “The surface on my Existing Grade layer” beats “my terrain.” Precise names are most of the specification.
- Clip to the site. If the surface runs past your property line, tell the agent to ignore faces outside the boundary curve on your Limit of Work layer.
Check. Pick three spots, measure the slope yourself between two contour points, and compare with the color the agent gave them.
T2
Layer Audit Against Your Standard RH
Save a copy of a messy project file. Paste your layer standard (your office’s, your studio’s, or your own) into the conversation, or point the agent to the file that holds it, and ask for a report before any change:
Here is our layer standard. Compare every layer in this document against it. List the layers that do not match, what you would rename or merge each one to, and any objects on layer 0. Change nothing yet.
Read the list. Correct what it got wrong. Then:
Apply the changes you listed, except the ones I crossed out. Delete no objects. Report what you did.
Check. Compare the object count before and after; it should not change. Open the layer panel and spot-check five layers against the standard.
T3
Block Count to a Plant Schedule RH
In a planting plan where each species is a block:
Count every block instance by block name, grouped by layer. Write a CSV plant schedule with columns for symbol, block name, quantity and layer, save it next to this file, and show me the table.
If your blocks carry botanical names or sizes as attributes, ask the agent to include them.
Check. Pick one species and count it yourself with SelBlockInstanceNamed. The numbers should match. If they do not, ask the agent why before you trust the rest.
+
Stretch: a Grasshopper Definition You Can Read GH
Ask the agent to turn T1 into a Grasshopper definition with sliders for the thresholds, with the code kept in a separate .py file next to the .gh. Then open the .py and read it. You will not follow every line; you will see where the thresholds live and what the script does with them.
What You Installed
Two halves. The MCP server is a small program your agent starts; it tells the agent which Rhino tools exist. The plug-in is loaded inside Rhino; mcpstart opens its door. Neither works alone, which is why the connection drops when Rhino closes. Expect this shape for every program you connect next: an adapter outside, and a way in.
Before You Use It on Real Projects
- Work on copies. Save As before each session. The agent edits the open document.
- Know where the data goes. What the agent reads from the model is sent to the AI provider. Check the provider’s terms, and any client contract or school policy, first.
- A person checks it. Treat every result as a draft, looked over before it goes out.