MCP (Model Context Protocol) server for Word document operations - read, write, edit, and manipulate .docx files with AI assistants
Server has 6 tools with basic schemas and descriptions in Chinese. All tools have input schemas with type definitions and descriptions present. However, descriptions are minimal (avg ~80 chars, well below the 194-char baseline for A+ tools), lack actionable context for LLM selection, and omit critical details like when to use each tool vs. alternatives. Parameter descriptions are present but terse. Output schemas are not documented, tools return plain strings rather than structured JSON. No error handling guidance; errors are raw exception messages. Tool naming is clear (verb_noun pattern) but compositions lack idempotency guarantees and chaining metadata. No tool annotations (readOnlyHint, destructiveHint). Security: file_path parameters accept user input without path traversal validation, critical for a file manipulation tool.
向文档追加带有特定样式的段落。支持设置中文字体(如宋体/黑体)。
创建一个新的空白 Word 文档。
获取 Word 文档的基本信息概览,包括段落数、表格数、各段落的简要信息。用于在读取完整内容前快速了解文档结构。
分段读取 Word 文档的内容和样式信息。支持分页读取大文档。
读取 Word 文档中的表格内容。
替换指定段落的内容,并应用新的格式。
Output schemas not documented. All tools return plain strings instead of structured JSON. LLMs cannot parse results reliably or chain outputs to downstream tools.
Path traversal vulnerability. file_path parameters accept user input without validation. Malicious input like '../../../etc/passwd' could read/write arbitrary files.
Descriptions lack actionable context. Avg 48 chars vs. 194-char baseline. Missing: when to use each tool, prerequisites, what to do if it fails. LLMs cannot reliably select the right tool.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | <=2025-11-25 | v2 |
No error handling guidance. Errors return raw exception strings (e.g., 'Error creating document: [exception]'). LLMs cannot determine if error is retryable, user-fixable, or fatal.
No tool annotations. Destructive tools (create_new_document, add_formatted_paragraph, replace_paragraph) lack destructiveHint. Read-only tools lack readOnlyHint. LLMs cannot infer safety properties.