MCP server for OpenSCAD — parametric CAD generation for laser cutting, 3D printing, and CNC
Server has 6 tools with complete input schemas and descriptions, but quality is inconsistent. Tool names follow verb_noun pattern (generate_*), which is good. Descriptions are present but generic (avg ~80 chars, below 194-char baseline). Parameters have types and descriptions, but lack enums for constrained inputs (e.g., lid_type, output_format accept free strings). No output schemas documented. Error handling exists (validate_config_parameters) but error messages are not recovery-focused. Security checks are present (banned keywords, path validation) but not reflected in tool descriptions. Overall: functional but below production baseline.
Generate a parametric 3D-printable box with optional lid
Generate a parametric laser-cut box with optional dividers and lid
Generate a custom OpenSCAD model from user-provided code
Generate a finger joint test pattern to calibrate finger dimensions
Generate a kerf test pattern to calibrate laser cutter
Generate a parametric laser-cut part with finger joints and optional openings
No output schemas documented. Tools return file paths and generated models, but LLMs cannot plan downstream actions without knowing response structure (e.g., does it return file_path, file_content, preview_url?).
Constrained parameters lack enums. 'lid_type' accepts 'snap'|'slide'|'none' and 'output_format' accepts 'svg'|'stl'|'png', but are defined as free strings. LLMs will hallucinate invalid values.
Tool descriptions are generic and under 100 chars. E.g., 'Generate a parametric laser-cut part with finger joints and optional openings' does not explain WHEN to use this vs generate_box, or what the output is used for.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Error messages in validate_config_parameters are actionable but not returned to LLM. Validation happens server-side; if a call fails, the LLM receives a generic error without guidance on what to fix.
Security constraints (banned keywords, file-reading function blocks) are enforced but not documented in tool descriptions. LLMs cannot understand why certain inputs fail without explicit guidance.