Rhino MCP (McNeel) is the official MCP platform published by McNeel for connecting AI agents to Rhino 3D and Grasshopper. The mcneel/RhinoMCP repository presents it as an MCP server for agents to create and edit inside Rhino. During the GitHub check on August 26, 2026, the repository was MIT licensed, had 274 stars and 40 forks, and its latest public release was Connector 0.1.5, published on July 16, 2026 with a connector.mcpb asset.
For Educasium, this page should be distinct from community RhinoMCP. The main signal here is not only a tool list: it is that Rhino's publisher provides architecture, documentation, a privacy policy, a desktop connector, and setup commands for Claude Desktop, Claude Code, GitHub Copilot, OpenAI Codex, Gemini CLI, and local models. That makes it a natural option to study for schools, studios, and firms that want to stay close to the McNeel ecosystem.
Contents
- What Rhino MCP (McNeel) does
- Official architecture
- Installation and supported assistants
- Verified Rhino and Grasshopper capabilities
- Privacy and security
- Comparison with community RhinoMCP
- Educasium position
What Rhino MCP (McNeel) does
Key point: Rhino MCP (McNeel) is the official initiative for driving Rhino and Grasshopper from MCP-compatible agents. It targets creation, editing, scripting assistance, and development workflows around Rhino.
A platform rather than a simple script
The README calls it the Rhino MCP Platform. The official documentation explains that any assistant speaking the Model Context Protocol can drive Rhino and Grasshopper, with separate guides for Claude Desktop, Claude Code, GitHub Copilot, OpenAI Codex, Gemini CLI, and local models. This is broader than a single Claude setup: McNeel documents multiple clients and an MCP router.
The desktop connector manifest describes it as an MCP bridge between Rhino 3D, Grasshopper, and Claude. It can be used as a modeling, scripting, or plugin development assistant. The same manifest says it requires the Rhino-MCP-Platform plugin, installable from Rhino's Package Manager.
Do not confuse it with the community project
The community jingcheng-chen/rhinomcp repository and the official mcneel/RhinoMCP repository have similar names, but they are different projects. The first is a third-party integration with a very detailed tool reference in its README. The second is published by McNeel and organizes its content around a connector, router, Rhino plugin, official documentation, and support through Discourse or GitHub issues.
For the Educasium reader, that distinction avoids a bad decision. An instructor can compare both, but should not automatically transfer capabilities, commands, or risks from one repository to the other. Each page needs its own sources.
Official architecture
Key point: McNeel's architecture combines a desktop connector, a router, a Rhino plugin, and shared layers for communicating with Rhino and Grasshopper. The repository is not just a hand-installed Python server.
Connector and router
The connector manifest declares a Node server with router-launcher.mjs as the entry point. Its MCP configuration launches node with the router, a default Rhino version argument, a startup timeout, and variables for locating Rhino 8, Rhino 9, or Rhino WIP. The manifest compatibility targets Claude Desktop >=1.0.0, darwin and win32 platforms, and Node >=18.0.0.
The README in the rhino folder says requests go through the router app rather than simple HTTP. It says this organization gives a more stable connection, avoids socket interference, lets the MCP server pick it up immediately, and can launch as many Rhino instances as needed. The rhino/router folder contains RhinoManager, RhinoLocator, SlotStore, SpawnDiagnostics, ProxyDispatcher, and SpawnSlotTool.
Rhino plugin and Grasshopper resources
The repository contains a rhino/plugin folder with commands such as MCPStartCommand, MCPConnectCommand, MCPSpawnCommand, MCPHelpCommand, AISettingsCommand, and AIPanelCommand. It also contains verified tools such as GetContextTool, GetViewportImageTool, ListObjectsTool, OpenDocTool, SaveDocTool, RunCommandTool, RunPythonTool, RunCSharpTool, SetCameraTool, SetSelectionTool, SetLayerMaterialTool, ZoomToLayerTool, and ZoomToObjectTool.
For Grasshopper, verified paths include GH1 and GH2 tools: start, search components, describe component, place component, place slider, connect, connect many, get canvas graph, apply graph, solve, and clear canvas. The troubleshooting docs say gh2 Grasshopper 2 tools require Rhino 9 WIP or BETA. That is an important limit to explain before any workshop.
Installation and supported assistants
Key point: McNeel documents several setup paths, but the simplest Claude Desktop path uses an mcpb connector and the Rhino-MCP-Platform plugin. For Codex and other clients, the MCP configuration points to the Rhino router.
Claude Desktop and connector
The README says the fastest path is Claude Desktop: install the Rhino3d connector and let it install the Rhino plugin. The verified manifest describes the connector as Rhino MCP Platform version 0.1.5. The latest release checked the same day is connector-v0.1.5 and contains connector.mcpb.
There is a small operational caution: the root README still pointed to a connector-v0.1.3 URL, while the newest release and inspected manifest indicated 0.1.5. For an Educasium tutorial, the correct instruction is therefore to check the Releases page at install time and not copy an old URL mechanically.
Codex, Copilot, Gemini, and local models
The Getting Started documentation offers several guides. For Codex, it explains installing the Rhino plugin, then running MCPConnect inside Rhino. That command prints the command Codex needs to launch the Rhino MCP router. The user then adds a rhino entry to the Codex configuration with the rhino-mcp-router command and a default version argument.
The same section documents GitHub Copilot, Claude Code, Gemini CLI, and LM Studio. For Educasium, this is useful because training can separate two skills: installing the connector in a given assistant and understanding what Rhino receives. The tool is not tied to a single chat interface.
Verified Rhino and Grasshopper capabilities
Key point: source files confirm tools for context, viewport, documents, selection, code, Rhino commands, camera, materials, and Grasshopper. File names give a concrete capability picture without promising more than the repository shows.
Rhino, documents, and viewport
The verified plugin-side tools cover context reading, object listing, selection, viewport, documents, and some scene actions. GetContextTool and ListObjectsTool help understand current state. GetViewportImageTool gives visual feedback. OpenDocTool and SaveDocTool touch files. SetCameraTool, ZoomToLayerTool, and ZoomToObjectTool affect the view. SetLayerMaterialTool indicates material or layer action.
That coverage supports training uses: request a simple scene, capture the result, adjust the camera, inspect objects, save a copy. It is not enough to guarantee a complete architecture or fabrication deliverable. As with any Rhino-linked tool, check dimensions, layers, units, tolerances, and domain intent.
Commands, Python, C#, and Grasshopper
RunCommandTool, RunPythonTool, and RunCSharpTool are the three sensitive tools to name clearly. The privacy policy also names them as tools that can execute arbitrary commands or code inside the Rhino process. They may be useful for scripting assistance and plugin development, but require control, review, and use on copies.
For Grasshopper, GH1 and GH2 files indicate base coverage: start, search for a component, describe a component, place components or sliders, connect multiple elements, get the canvas graph, apply a graph, solve, and clear the canvas. Advanced documentation also mentions migrating a GH1 script to GH2 and a sketch-to-model workflow, showing an orientation toward users who already know Rhino and Grasshopper.
Privacy and security
Key point: McNeel's privacy policy says the connector does not collect, log, or transmit user data on its own, but MCP responses go back to the AI client. That nuance is essential for responsible training.
Local processing and output to the assistant
The privacy policy, updated on May 19, 2026, describes the connector as a local desktop extension that bridges Claude to Rhino 3D and Grasshopper on the user's own machine. It says the connector does not phone home, emit telemetry, or maintain analytics.
The same policy explains that when a tool is invoked, the connector reads or modifies state inside the local Rhino process, then returns the result to the AI client. That means geometry, commands, script output, viewport screenshots, and similar content may leave the machine through Claude or the chosen assistant. The connector's policy does not replace the privacy policy of the model or client being used.
Files and arbitrary execution
The policy names open_doc and save_doc as examples of tools able to write files when the user asks through Claude. It also names run_command, run_python, and run_csharp as tools that can execute arbitrary commands or code inside Rhino. It says those tools can touch any file accessible to the user account.
That sentence is the core Educasium guardrail. In training, work on demonstration documents, keep sensitive client folders out of scope, show what goes back to the assistant, and explain how to stop an action. For a firm, this is a security decision, not just a creative tooling decision.
Comparison with community RhinoMCP
Key point: McNeel's option brings publisher proximity, official docs, and connector-oriented architecture; the community option brings a very detailed tool reference and simple uvx installation. Both deserve separate tests.
| Option | Natural use | Main strength | Limit to watch |
|---|---|---|---|
| Rhino MCP McNeel | Organizations that prefer publisher docs and official channels | Official connector, privacy policy, Discourse support | Connector/docs versioning should be checked at install time |
| Community RhinoMCP | Technical Rhino + Grasshopper training and prototyping | Very explicit tool reference in README | Third-party project not published by McNeel |
| Blender MCP | Free 3D introduction and prototyping | Accessible software and large community | Less natural for NURBS and Grasshopper |
| Houdini MCP | Procedural, wrangles, VFX, Houdini graphs | Advanced node-based control | More technical and less suited to classic Rhino design |
Decide by evidence, not by name
An official tool still needs testing. A good comparison runs the same task in both Rhino connectors: open a scene, create simple geometry, read objects, capture the viewport, build a small Grasshopper graph, save a copy, and observe errors.
Likewise, a community tool is not automatically inferior because it is unofficial. Community RhinoMCP may be more explicit about some tools. The sound Educasium method is to separate status, capabilities, security, installation, and observed results.
Educasium position
Key point: Educasium can recommend Rhino MCP (McNeel) as the official entry point to test for Rhino and Grasshopper, with strong guardrails around privacy and code execution. This is a framing page, not a production validation.
Responsible training
An Educasium workshop could follow the official docs: install the connector, install Rhino-MCP-Platform, run MCPConnect or the Claude Desktop path, open a clean document, request a simple box, capture the viewport, and inspect context. The Try It Out documentation specifically recommends starting with a fresh document and a simple shape to confirm that everything is wired correctly.
Then the workshop can expand into recipes: parametric furniture, spiral staircase, document audit, viewport capture, quick dimensions, Grasshopper from scratch, and explaining an existing definition. These examples are useful, but they should remain exercises. Client deliverables require measurements, domain validation, and data control.
Logical next step
The logical next step is an Educasium evaluation grid for Rhino connectors: Rhino 8 or 9 WIP compatibility, installation, available tools, data policy, code control, Grasshopper quality, visual feedback, crash recovery, and learner accessibility. That grid lets teams choose a connector by context instead of trusting only a demo video.