MCP server for Rhino 3D modeling application, providing tools for creating and manipulating geometric objects, managing scenes, and controlling Rhino via socket-based communication
RhinoMcpServer has moderate definition quality with some solid foundations but notable gaps. All 6 tools are explicitly registered with the [McpTool] attribute and have descriptions. Tool names follow verb_noun convention (create_*, get_*, clear_). Schemas are present with typed parameters and descriptions. However, output schemas are undocumented (all tools return plain strings instead of structured JSON), parameter descriptions lack format/range constraints, and error handling is generic. Color parameters accept free-form strings instead of enums, inviting hallucinated values. The server lacks pagination support, per-item error reporting, and recoverable error guidance. Average tool score is 57.7 across the 6 tools.
Clears all objects from the current scene
Creates a box with the specified dimensions
Creates a cylinder with the specified base point, height, and radius
Creates a new layer in the Rhino document
Creates a sphere with the specified center and radius
Gets information about objects in the current scene
Output schemas are undocumented. All tools return plain strings (e.g., 'Created sphere with ID: {id}') instead of structured JSON objects. LLMs cannot extract fields for downstream tool chaining, parse errors reliably, or build complex workflows. Example: create_sphere returns a string but agents need to extract the ID for subsequent operations.
Color parameter accepts free-form strings without enum constraint. Parameter description says 'Optional color for the sphere (e.g., red, blue, etc.)' which invites LLMs to hallucinate invalid color names or formats. Should use an enum: ["red", "blue", "green", "yellow", "cyan", "magenta", "white", "black"] or a regex pattern.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 53 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 43 | - | v1 |
Error responses lack recovery guidance. Generic try-catch blocks return unstructured error messages (e.g., 'Error creating sphere: {ex.Message}'). LLMs cannot determine if an error is retryable, user-fixable, or fatal. Example: 'No active document found' could guide users to 'Open a Rhino document first', but the current response gives no hint.
Parameter descriptions lack format and constraint documentation. Numeric parameters (centerX, centerY, centerZ, radius, width, depth, height, baseX, baseY, baseZ) have no documented ranges, units, or minimum/maximum values. LLMs may pass zero, negative, or extremely large values that cause silent failures or API errors.
Destructive tools (clear_scene) lack confirmation or dry-run support. The clear_scene tool can delete all scene objects with no warning or undo path. Agents making mistakes will irreversibly destroy user work. Should implement a confirmation pattern or return a 'confirm required' response.
get_scene_info has no documented return schema. The description says 'Gets information about objects in the current scene' but does not specify what fields are returned, how many objects, whether pagination applies, or the structure of each object. LLMs cannot plan downstream operations.
No pagination support. If the scene has hundreds of objects, get_scene_info will return all of them, blowing the context window. Tools returning lists should accept page/offset and limit parameters and return a total count.
Tool descriptions do not indicate state-changing side effects. clear_scene, create_sphere, create_box, create_cylinder, and create_layer all modify the Rhino document, but their descriptions do not explicitly state this. Agents need to know which tools are safe to retry and which have irreversible consequences.
Layer name parameter in create_layer lacks format validation. Should accept layer names 1-255 characters, alphanumeric with hyphens/underscores, no leading/trailing spaces. Current description just says 'Name of the new layer' with no constraints.