MCP server that bridges Claude to WinDbg GUI via DbgEng COM interface, enabling remote debugging commands and crash analysis
Server has 5 tools with visible schemas and descriptions, but quality is significantly below production baseline. All tools lack English descriptions (descriptions are in Chinese), making them inaccessible to most LLM agents. Parameter descriptions are minimal or missing. No output schemas documented. Error handling is generic ('ERROR: {ex.Message}'). Tool names use non-standard prefixes ('wd_' instead of verb_noun). No input validation or recovery guidance. Schemas are present but sparse, many parameters lack descriptions entirely (e.g., 'thread' in wd_callstack has no description). This server would not pass code review for production use.
自动执行完整 crash 分析:!analyze -v + 调用栈 + 故障模块信息
获取当前线程调用栈
在 WinDbg GUI 中执行调试命令。命令会真正在 GUI 里执行,你可以实时看到输出。
智能检查:给定地址或符号,自动选择 dt / dps / da / u 显示内容
获取调试会话状态(执行状态、目标类型)
All tool descriptions are in Chinese, not English. LLM agents trained on English cannot understand when or why to call these tools. Descriptions must be in English for cross-language LLM compatibility.
Tool names use non-standard 'wd_' prefix instead of verb_noun pattern (e.g., 'execute_command', 'analyze_crash', 'get_callstack'). LLMs rely on action verbs to infer intent. Current names are opaque.
No output schemas documented. Callers cannot predict what fields to expect from tool responses. wd_execute, wd_analyze, wd_callstack, wd_inspect return raw strings; wd_status returns unstructured text. LLMs cannot plan downstream operations without knowing response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | 2026-07-28+ | v2 |
Error handling is generic and non-actionable. Catch block returns 'ERROR: {ex.Message}' with no recovery guidance. LLMs cannot determine if errors are retryable, user-fixable, or fatal. No error classification or suggestions for next steps.
Parameter descriptions are missing or trivial. 'thread' in wd_callstack has no description. 'hint' in wd_inspect lacks explanation of valid values (struct/string/code/raw). 'timeout' lacks range constraints. LLMs cannot validate inputs without explicit parameter documentation.