MCP server that bridges AI assistants to SketchUp via HTTP. Embeds a lightweight TCP/HTTP server inside SketchUp so that an external MCP server (Python) can drive SketchUp programmatically.
Server has 25 tools with complete input schemas and descriptions. Naming follows verb_noun convention (get_, list_, create_, delete_). However, descriptions are often terse (10-50 chars), parameter descriptions are minimal or absent, and output schemas are not documented. Error handling is basic. No tool annotations (readOnlyHint, destructiveHint). Parameter validation rules are not stated in descriptions. The Ruby plugin implementation is visible but tool definitions are inferred from the Python server code rather than explicitly registered with full metadata.
Create an arc (as a series of edges) with a given center, radius, and angular range.
Create a 3-D box (rectangular prism) by specifying width, depth, height. Optionally provide an origin [x, y, z] for the base corner.
Create a circle (as a series of edges) at a given center, radius, and normal direction.
Create a line segment (edge) between two 3D points.
Create a flat face (polygon) in SketchUp from an ordered list of 3D points. Points are in SketchUp's default unit (inches unless model is metric).
Create an empty named group in SketchUp, optionally on a layer.
Create a regular polygon (N-sided) with a given center, radius, and number of sides.
Descriptions are terse (10-50 chars) and lack context for LLM tool selection. E.g., 'List all layers (tags) in the SketchUp model with their visibility.' does not explain WHEN to call this vs other discovery tools or what the agent should do with the result.
No output schemas documented. Tools return structured data (e.g., create_face returns {id, type, area, normal}) but LLMs cannot see the response structure in advance. This forces agents to guess what fields are available and risks failed downstream tool calls.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot distinguish safe read-only tools from destructive ones (delete_entity, save_model) without explicit hints. This risks accidental data loss.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Delete an entity (face, edge, group, component instance) by its ID.
Apply the Follow Me operation: extrude a face along a path of edges.
Get the length of an edge.
Get the bounding box of an entity (min and max corners).
Get the area of a face.
Get the normal vector of a face.
Get general information about the currently open SketchUp model: name, file path, unit system, entity/layer/material counts.
List all component definitions in the SketchUp model.
List entities (faces, edges, groups, component instances) in the model or inside a named group.
List all layers (tags) in the SketchUp model with their visibility.
List all materials in the SketchUp model with their colors and textures.
Extrude a face by a given distance (positive or negative).
Redo the last undone operation.
Rename a group or component instance.
Save the current SketchUp model to disk.
Assign an entity to a layer (tag). Creates the layer if it does not exist.
Assign a material to an entity. Creates the material if it does not exist.
Undo the last operation.
Parameter descriptions are missing or minimal. E.g., create_face 'layer' param has description 'Layer name to assign (created if missing)' but does not explain what a layer is, when to use it, or valid naming constraints. Agents cannot reason about when to pass this parameter.
Error handling is not visible in tool definitions. No guidance on retryability, user-fixable errors, or recovery steps. E.g., if create_face fails due to invalid points, the agent has no hint to validate input or ask the user for clarification.