Universal API Connector MCP for AI applications - manages API configurations, fetches data, stores results, and provides data preview capabilities
Three tools with partial schemas and descriptions. manage_api_config and fetch_api_data have action/enum-like parameters but lack explicit enum constraints in schema. Descriptions are present (50-100 chars) but generic and lack WHEN/WHY guidance. api_data_preview has 10 parameters but many lack actionable constraints (max_rows, max_cols, preview_depth are unbounded integers). No output schemas documented. Error handling returns JSON but lacks recovery guidance. Tool names use underscores (good) but descriptions rely on emoji rather than clarity. No tool annotations (readOnlyHint/destructiveHint). Schemas visible but incomplete, parameters lack minima/maxima and enum constraints.
👁️ API数据预览工具 - 预览API数据而不存储
📡 API数据获取工具 - 从API获取数据并自动存储
🔧 API配置管理工具 - 管理API连接配置
No output schemas documented. LLMs cannot infer what fields to expect from tool responses, forcing unstructured parsing and risking context loss.
Numeric parameters (max_rows, max_cols, preview_depth, truncate_length) lack min/max bounds. LLMs can pass absurd values (max_rows=999999) causing timeouts or memory exhaustion.
action parameter in manage_api_config lacks enum constraint in schema. Free-form string invites hallucinated values like 'remove' or 'check' instead of documented list/add/update/delete/validate/test.
Descriptions are emoji-heavy and lack WHEN/WHY context. '🔧 API配置管理工具 - 管理API连接配置' (Chinese) does not explain when to call this vs other tools or what prerequisites exist.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | <=2025-11-25 | v2 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot distinguish safe reads from destructive writes. fetch_api_data and manage_api_config (add/update/delete) should be marked destructive.