Loading ToolBench
MCP server for controlling Blender with AI — 359 tools across modeling, animation, rendering, and more
The Blender MCP server presents a specialized, focused tool collection with structured schemas and reasonable descriptions. All 12 tools shown have input schemas with proper JSON Schema format (types, descriptions, enums where appropriate). Tool naming follows verb_noun patterns consistently (blender_uv_*, blender_texture_*). However, descriptions are terse (averaging ~30-50 chars) and lack context on WHEN to use each tool vs. alternatives, preventing context window optimization. Parameter descriptions are present but minimal. Output schemas are not documented, LLMs cannot predict response structure. Error handling and recovery guidance are absent from visible code. The server is STDIO-only, which is a hard cap at 50 for protocol readiness. Definition quality is solid but not exceptional; descriptions need expansion to match production baselines (194 chars average).
Create a texture atlas from multiple objects.
Analyze UV layout quality and detect issues.
Automatically generate UV seams based on hard edges and angles.
Apply a checker grid texture to visualize UV layout.
Normalize UV density across objects for consistent texel density.
Optimize UV layout for production (margin, stretch, packing).
Optimize UV island layout.
Tool descriptions are too short (30 - 50 chars average vs. 194 char baseline). Lack context on WHEN to use each tool, prerequisites, and differentiation from similar tools. E.g., 'Automatic UV unwrap.' does not explain when to choose this over smart_project or project.
Output schemas are not documented anywhere in the visible code. LLMs cannot predict response structure, field names, or types. This forces agents to guess what the tools return and risks context loss or hallucinated field references.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | <=2025-11-25 | v2 |
UV projection mapping (cube, cylinder, sphere).
Mark or clear UV seams.
Smart UV projection (automatic island splitting).
Transform UV coordinates (translate, rotate, scale).
Automatic UV unwrap.
No error handling guidance visible in tool descriptions. Tools do not indicate what can go wrong (e.g., 'Object not found', 'Invalid mesh topology'), what the error means, or how the agent should recover. This violates recovery-guide pattern.
Parameter descriptions are minimal (often 1 - 2 words: 'Object name', 'Margin'). Do not explain units, ranges, or consequences. E.g., 'margin' in blender_uv_pack lacks explanation of what value range is sensible or what 0.001 vs 0.1 means visually.
No documentation of which tools are read-only vs. destructive/write operations in descriptions. blender_uv_analyze is marked as READ_ONLY in metadata, but most tools are WRITE with no guidance on side effects, reversibility, or confirmation requirements.
Parameter naming is generic ('object_name') rather than hinting at acceptable formats. No guidance on whether to pass mesh name, object name, collection, or ID. Should clarify 'object_name (e.g., "Cube.001") must be the exact Blender object name visible in the outliner'.