11 tools (scene.get_summary, object.create_cube, object.create_plane, object.create_uv_sphere, object.set_transform, object.add_modifier_bevel, object.add_subdivision_modifier, material.create_principled, material.set_base_color, render.set_engine_cycles, render.set_resolution, render.render_still, list_objects, get_scene_info) have empty input schemas {} with no parameter definitions. This blocks LLMs from understanding what parameters to pass.

14 tools have descriptions under 30 characters (e.g. 'Create plane' for object.create_plane, '创建材质' for material.create_principled, '获取场景摘要信息' for scene.get_summary). These are too sparse for LLM selection and explain neither WHEN to use the tool nor WHAT happens to state. Animation tools are well-described (200-400 chars); these are not.

No output schemas documented for any tool. LLMs cannot plan downstream steps without knowing what fields will be returned (e.g., after list_objects, what fields identify each object? Is it 'name', 'object_name', or 'id'?). This server has 0%.

Parameter descriptions missing for 12 tools with partial schemas (create_primitive, delete_object, get_object_info, transform_object, set_material). E.g., transform_object has 'location' and 'rotation' params but does not document their format (are they [x, y, z] arrays? degrees or radians?).

No error recovery guidance in tool descriptions. Animation tools return _result(False, None, str(e)) but the spec does not document error cases or how LLMs should respond. E.g., if anim_add_uv_scroll fails because the material does not exist, should the LLM create it first? Fall back to a different material? Delete keyframes?

Tool naming is inconsistent across the suite. Some use verb_noun (create_primitive, list_objects, delete_object), others use module.verb (object.create_cube, material.create_principled), and some use verb_adjective_noun (anim_add_uv_scroll). Recommendation: standardize on verb_noun_detail (e.g., create_cube, add_bevel_modifier, set_cycles_engine).

Tool descriptions in Chinese may not be parsed correctly by English-optimized LLMs. While not a hard violation, descriptions like '为 Mapping 节点的 Location 添加基于帧的 Driver，实现 UV 滚动动画' force LLMs to translate on the fly, introducing latency and potential misinterpretation. Recommend English descriptions for all tools serving diverse LLM backends.

No tool annotations (readOnlyHint, destructiveHint, idempotentHint) declared in schemas. delete_object and render operations are destructive but unmarked. Recommend adding destructiveHint to delete_object and render.render_still.