How to install Rhino MCP for Claude, Codex, and other AI assistants

01

Confirm the official integration

Rhino MCP is McNeel's free MCP platform for connecting Rhino and Grasshopper to supported AI assistants. McNeel currently documents setup paths for Claude Desktop, Claude Code, GitHub Copilot, OpenAI Codex, Gemini CLI, and local models.

Confirm that you are using the official McNeel documentation and repository before installing anything. Record your Rhino version, operating system, selected AI client, and the project copy you will use for testing.

• — McNeel documentation
• — Supported AI client
• — Rhino version
• — Disposable test file

02

Install for the AI client you actually use

Open McNeel's Getting Started path and choose your AI assistant. Follow that client-specific connector or configuration process, then complete the Rhino-side plug-in installation described by McNeel. Restart the applications if the guide requires it.

Do not copy a configuration from an unrelated client: desktop apps, command-line agents, editors, and local models can store MCP settings differently.

03

Verify the connection with a reversible test

Open a blank or copied Rhino file and begin with a small inspection request. Confirm that the assistant can identify the active document and selected objects before allowing it to create geometry or run commands.

Then test one reversible action, inspect the result in Rhino, undo it, and review any log or tool-call record your client exposes. This establishes that the connection, units, selection context, and rollback path work.

• — Read current document state
• — Run one bounded action
• — Inspect geometry and units
• — Undo successfully

04

Do not confuse Rhino MCP with Raven

Rhino MCP is an MCP connection that lets an external AI assistant work with Rhino and Grasshopper. Raven is a separate Rhino 8 plug-in installed through Rhino's Package Manager and used inside Rhino and Grasshopper. They are different products and should not be described as the same installation path.

Choose the integration that fits the intended client, data boundary, and workflow. If a Prompt references either tool, it should name the exact dependency and tested version.

05

Add project safeguards before real work

For production use, work from versioned project copies, limit each instruction to a named object set, define units and tolerances, and require a preview or verification step before baking, deleting, exporting, or overwriting work.

A successful connection proves that tools can communicate. It does not prove that a generated model operation is correct, coordinated, buildable, or appropriate for the project.