Loading ToolBench
MCP server for AutoCAD integration with database support for entity management and text pattern analysis
This AutoCAD MCP server has significant quality gaps that would prevent reliable use in production. While 6 tools are explicitly defined with visible schemas and descriptions, the descriptions are in Chinese (limiting LLM comprehension for English-focused agents), parameters lack proper constraints, output schemas are not documented, and error handling provides minimal guidance for agent recovery. The server also uses STDIO transport, which is not remotely accessible. Tool naming is reasonably verb-first (create_new_drawing, draw_line, draw_circle, scan_all_entities, highlight_entity, count_text_patterns), but schema quality is inconsistent, some tools lack proper parameter descriptions or type information in the structured format.
统计图纸中文本实体中特定模式的出现次数 Args: pattern: 要搜索的文本模式，默认为"PMC-3M"
创建新的 AutoCAD 图纸
在AutoCAD中绘制圆 Args: center_x: 圆心X坐标 center_y: 圆心Y坐标 radius: 半径 layer: 可选的图层名称
在AutoCAD中绘制直线 Args: start_x: 起点X坐标 start_y: 起点Y坐标 end_x: 终点X坐标 end_y: 终点Y坐标 layer: 可选的图层名称
通过Handle在AutoCAD中高亮显示指定实体 Args: handle: 实体的Handle值 color: 高亮颜色码（1=红色, 2=黄色, 3=绿色, 4=青色, 5=蓝色, 6=洋红色）
扫描当前图纸中的所有实体并保存到数据库
All tool descriptions are in Chinese, limiting comprehension for English-fluent LLMs and causing potential tool selection confusion in multilingual agents. The MCP spec and Arcade tool patterns expect descriptions in English (or at minimum, the UI language of the target LLM).
No output schemas are documented for any tool. LLMs cannot plan downstream operations or extract fields if they don't know what the response structure is. This violates the requirement that 100% of A+ tools have documented return types.
Parameters lack proper constraints. Numeric parameters (start_x, start_y, end_x, end_y, center_x, center_y, radius) have no min/max bounds documented. The color parameter in highlight_entity includes valid values (1-6) only in the description text, not as a formal enum constraint in the schema.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 47 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
scan_all_entities has no pagination or result limiting. A drawing with thousands of entities would return all of them, potentially exhausting the LLM context window. The baseline pattern requires limit and offset parameters or a documented max result count.
Error messages are generic string concatenations (e.g., 'Failed to create line: {error}') with no recovery guidance. Agents cannot determine whether to retry, ask the user, or abandon the operation. Missing: error classification and actionable next steps.
Tools that modify state (create_new_drawing, draw_line, draw_circle, highlight_entity) lack confirmation or dry-run steps. An agent could inadvertently create many drawings or highlight wrong entities without a safety gate.
Parameter descriptions lack operational detail. E.g., 'layer' parameter says '可选的图层名称' (optional layer name) but does not specify: Does the tool create the layer if it doesn't exist? Can layer names contain spaces or special characters? What is the naming convention?
create_new_drawing accepts a 'template' parameter but provides no guidance on valid template paths, where templates are stored, or error handling for nonexistent templates. This forces agents to guess or fail and retry.