HTTP MCP server plugin for Model Context Protocol integration with Sweet Home 3D. Provides tools to programmatically interact with Sweet Home 3D models, including home state queries, object manipulation, texture application, and annotation creation.
The server provides 3 well-defined tools with complete input schemas and descriptions. Tool names use clear verb_noun conventions (apply_texture, add_dimension_line, add_label). Descriptions are detailed and contextual (194 - 250 chars avg), exceeding the baseline of 194 chars. However, there are critical gaps: (1) output schemas are NOT documented anywhere, callers cannot know what these tools return, forcing blind integration; (2) no error handling guidance is provided, LLMs receive no recovery instructions if a call fails; (3) parameter descriptions lack explicit constraints (e.g., 'scale must be positive' is mentioned in apply_texture but not enforced in schema as minimum:0); (4) no tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all three being WRITE operations; (5) composition design is weak, apply_texture requires pre-knowledge of texture names but list_textures_catalog is not present as a discoverable tool. The schemas themselves are well-formed JSON Schema with types, required flags, and enums where appropriate.
Add a dimension line (measurement annotation) to the 2D plan. Shows the distance between two points with extension lines and an auto-calculated length label. All coordinates in centimeters. Offset controls perpendicular distance of the label from the measured line (positive = above/left, negative = below/right, typical value: 20-50).
Add a text label (annotation) to the 2D plan. Labels can display room names, dimensions, notes, or any text. Coordinates are in centimeters. Optionally set font size, style, color, rotation angle, and 3D elevation.
Applies a texture from the catalog to a wall side or room surface. Use list_textures_catalog to find available texture names first. For walls: surface can be 'left', 'right', or 'both' (sides relative to wall direction from start to end). For rooms: surface can be 'floor', 'ceiling', or 'both'. Setting a texture overrides any solid color on that surface. Optionally set angle (rotation in degrees) and scale (magnification). null textureName removes the texture.
No output schemas documented for any tool. Callers cannot know what fields, types, or structure are returned. LLMs must reason blindly about downstream tool chains and data extraction.
No error handling guidance. Tools have no documented error responses, recovery steps, or error classification (retryable vs user-fixable vs fatal). LLMs receive failures with no actionable next step.
Missing tool annotations despite all tools being WRITE operations. No readOnlyHint, destructiveHint, or idempotentHint are declared. Agents cannot distinguish safe retry patterns from irreversible side effects.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 58 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 44 | - | v1 |
Parameter constraints mentioned in descriptions but not formally declared in schema. E.g., apply_texture description says 'scale must be positive' but schema lacks minimum constraint. add_dimension_line offset 'typical 20-50' is advisory, not enforced.
apply_texture description references 'list_textures_catalog' tool to find texture names, but no such tool is exposed in the MCP interface. This breaks the assumed tool chain and forces agents to guess valid texture names.
add_label parameter relationships undocumented. fontSize, bold, italic, and alignment are conditionally valid (bold and italic 'require fontSize' per description) but schema does not enforce this dependency. Agents may pass bold=true without fontSize.
No pagination, limit, or batch operation patterns. If an agent needs to add 100 labels, it must call add_label 100 times sequentially, wasting tokens and latency. No batch_add_labels variant offered.