1. Home
2. Rhino MCP

Rhino MCP: connecting an AI agent to Rhino

Rhino MCP is a Model Context Protocol server that lets an AI agent read and edit a Rhino model — creating geometry, moving objects, managing layers and blocks through plain-language instructions. Unlike SketchUp, Rhino has an official, free, open-source option from McNeel, and for many people that is the right answer.

Updated 11 September 2026 · ZirenAI

What is Rhino MCP?

The Model Context Protocol (MCP) is an open standard for connecting AI agents to external tools. A Rhino MCP server exposes Rhino's modelling operations as typed tools an agent can call, so an assistant can inspect what is in a model and change it rather than guessing at RhinoScript.

The setup has two halves: a plug-in running inside Rhino that executes commands against the live document, and an MCP server process that translates the agent's tool calls into those commands. Because the agent calls typed operations instead of writing raw script, results are more predictable and failures are easier to read.

What are the options, and how do they compare?

Three routes exist as of September 2026.

McNeel publishes RhinoAI at github.com/mcneel/rhinoai under an MIT licence. It lists Claude Desktop, Claude Code, GitHub Copilot, OpenAI Codex, Gemini CLI and local models as compatible clients. The older mcneel/RhinoMCP repository now redirects there.

Should I just use McNeel's RhinoAI instead?

Often, yes — and we would rather say so than have you find out after paying us.

RhinoAI is official, MIT-licensed, runs against the Rhino on your own machine, and its connector installs the plug-in for you. The plug-in costs nothing; the AI client you point it at is your own to pay for. If you already pay for Claude, Copilot or Codex and you are comfortable pointing an MCP client at a server, it does the core job and costs you nothing extra. That is a genuinely good position for Rhino users to be in, and it is better than what SketchUp users have.

ZirenAI is worth paying for in three situations. First, if you do not want a second AI subscription to manage — model usage is paid for with points in the client. Second, if you want the connection looked after: we check it before every run and tell you which part failed rather than handing you a stack trace. Third, if your work is architecture or interiors — our prompts and tool sequences are tuned for layer standards, block management and floor-plan work rather than general modelling.

If none of those apply to you, use RhinoAI. It is the honest answer.

What about Grasshopper?

Several community projects target Grasshopper specifically, generating and rewiring definitions from plain language. ZirenAI does not touch Grasshopper definitions — it works on Rhino document geometry, layers and blocks, and leaves your definitions alone. If parametric definition authoring is what you are after, the Grasshopper-focused community servers are the right tool.

How do I set up Rhino MCP?

For the community servers you install the plug-in through the Rhino Package Manager, register the MCP server with your client in a JSON configuration file, and make sure your Python version matches what the server expects. RhinoAI's connector handles most of this for you.

1. Install ZirenAI. It finds the modelling software you have and sets each one up — plug-in, runtime and bridge — one click per application.
2. Pick Rhino in the app. The connection is checked automatically; if it fails you are told which part failed and how to fix it.
3. Describe the work. Plain language, plus a plan or reference image if you have one.

Full setup guide → · If the connection fails →

What can Rhino MCP actually do well?

Geometry that is 80% right is often worse than no geometry, because cleaning it up takes longer than modelling it yourself. The work worth handing over is the work where 80% is still useful:

• Layer, naming and block cleanup against an office standard
• Arraying and spacing — façade panels, seating, structural bays
• Turning a plan into a massing model — walls, slabs, openings
• Batch operations across many objects — renaming, re-layering, material assignment
• Auditing a file — counting, finding stray geometry, listing block definitions
• Preparing a model for export or for drawing layout

It is a poor fit for Class-A surfacing, topology decisions on complex NURBS, and anything that ends up on a stamped drawing. It moves geometry; it does not have taste.

Which Rhino versions are supported?

ZirenAI supports Rhino 7, 8 and 9 on Windows 10 and 11. macOS is in testing. McNeel's RhinoAI currently develops against a Rhino 9 branch; Rhino 9 entered public beta in late June 2026 and is open to Rhino 8 licence holders.

Common questions