Mobile automation testing via Model Context Protocol. Provides a standards-based interface for AI assistants to perform automated testing on Android and iOS devices using Appium.
This server has 13 tools with moderate quality. All tools have names that start with action verbs (list_, connect_, disconnect_, get_, find_, click_, input_, take_, swipe_) which is good. Descriptions exist for all tools but are in Chinese and lack depth about WHEN to use them or what downstream dependencies exist. Parameter schemas are present and mostly complete with type definitions, but descriptions are minimal (often just the parameter name translated). Output schemas are not documented anywhere in the visible code. No error handling guidance is provided, tools do not indicate what errors are retryable, what recovery steps are available, or how to handle failures. The server lacks composition guidance (e.g., which tools should be called before others). Critical issue: tool descriptions are in Chinese, which may confuse English-speaking LLM clients. See tool-by-tool scores below.
手动清理过期的设备会话
点击指定的UI元素
连接到指定的移动设备并创建会话
断开指定设备的连接并关闭会话
查找单个UI元素
获取指定设备的详细信息
获取指定会话的详细信息
在指定的UI元素中输入文本
列出所有可用的移动设备(Android和iOS)
All tool descriptions are in Chinese (e.g., '列出所有可用的移动设备(Android和iOS)'), making them unintelligible to English-speaking LLM clients. Descriptions must be in English for international use.
Parameter descriptions are minimal and often under 20 characters (e.g., '定位器类型' = ~10 chars, '会话ID' = ~6 chars). These provide insufficient context for LLM parameter selection.
No output schemas are documented in the visible code. Tools return Dict[str, Any] with no explicit schema definition. LLMs cannot plan downstream calls or understand what fields to extract.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 43 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 29 | - | v1 |
列出所有活动的设备会话
手动刷新和重新发现移动设备
在设备屏幕上执行滑动操作
对移动设备屏幕进行截图
No error handling guidance. Tools do not document what errors may occur, whether they are retryable, or what the LLM should do next. e.g., connect_device may fail if device is offline, but there is no recovery hint.
Parameters like 'locator_type' and 'locator_value' lack enums or format specifications. Valid locator types (id, xpath, accessibility id, etc.) are not constrained, inviting hallucinated values from the LLM.
tools like list_sessions and list_devices do not document pagination. If many sessions/devices exist, the response could be huge, exhausting context. No mention of limits or next_cursor.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are visible in the code. Some tools like cleanup_sessions and refresh_devices are destructive but are not marked as such, preventing LLM safety filters.
Tool composition is unclear. The sequence of operations (list_devices → connect_device → find_element → click_element) is not documented. Agents may not know which tools to call in what order.