加密环境文件操作 MCP 工具。当 Node.js 是加密软件白名单进程时,通过 fs 模块自动解密读写文件明文,替代 AI Agent 内置文件工具,适用于任何支持 MCP 协议的 Agent。
Strong schema definitions with comprehensive parameter constraints (min/max, enums, patterns). All 10 tools have descriptions and input schemas. Tool names follow verb_noun convention (read_file, write_file, edit_file, etc.). However, descriptions are terse (avg ~80 chars, below 194-char baseline) and lack actionable context for LLM selection. Parameter descriptions are present but minimal. Error handling is structured (result() wrapper with code/ok/warnings) but recovery guidance is sparse. No output schema documentation visible in tool registration. Security considerations (path traversal, file access control) are implemented but not documented in descriptions.
复制单个文件或目录树,支持跨分区和加密环境;失败保留源。
删除文件或目录树;支持dryRun预览;失败时保留源。
单次或批量编辑互斥;全部匹配成功后暂存提交。支持dryRun预览、expectedHash/expectedMatches、换行兼容和限时正则。
列举目录内容,支持glob过滤和递归;返回文件类型、大小和修改时间。
移动或重命名文件/目录,支持跨分区;失败保留源。
严格读取UTF8;最多40万字符,完整读取时返回hash,截断时返回nextOffset。
字符/行分页。offset按去BOM后的UTF16单元计算,请使用nextOffset;行模式流式读取并返回nextLine。
批量严格读取UTF8,最多100个文件,总内容预算40万字符,逐项返回状态。
Descriptions lack LLM-optimized context. Most descriptions (avg ~80 chars) are technical specifications rather than selection guidance. E.g., 'read_file' says 'strictly read UTF8; max 400k chars' but doesn't explain when to use it vs read_file_partial or when it's appropriate for the agent's task.
No output schema documentation. Tools return structured results via result() wrapper with {ok, code, changed, data, warnings}, but this schema is not exposed in tool registration. LLMs cannot plan downstream calls or validate response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2025-06-18+ | v2 |
按glob或正则搜索文件内容,返回匹配行及上下文;支持dryRun和限时计算。
写入或追加UTF8,暂存校验后提交,失败保留原文件;支持BOM/EOL、hash冲突检查、禁止覆盖和显式策略。
Error recovery guidance is minimal. Errors return code + message but rarely suggest next steps. E.g., 'READ_ONLY' error says 'service running in read-only mode' but doesn't guide the agent on what to do (retry elsewhere, ask user, etc.).
Parameter descriptions are minimal (avg ~30 chars). E.g., 'path' is described as '路径;相对路径以 MCP_BASE_DIR 或服务启动目录为基准' (Chinese; ~50 chars) but lacks examples of valid formats or constraints on special characters. LLMs cannot infer whether '.' or '..' are allowed.
Destructive operations (delete_file, move_file, copy_file with overwrite) lack confirmation/dry-run guidance in descriptions. dryRun parameter exists but is not highlighted as a safety mechanism in the tool description.