Loading ToolBench
Full-featured Markdown and document processing service for AI Agents (Claude Desktop, Cursor, Antigravity, VSCode, GitHub Copilot) via Model Context Protocol (stdio). Provides tools for fixing Markdown, converting documents to Markdown, web scraping, OCR, export, LaTeX processing, BibTeX parsing, and AI-assisted document workflows.
ReadMD MCP server has 19 tools with significant quality gaps. All tools have descriptions (positive), but most descriptions are in Chinese and lack English clarity for LLM reasoning. Input schemas are present but sparse: most tools have only 2-3 parameters with minimal type information. Critical issues: (1) No output schemas documented anywhere, LLMs cannot plan downstream calls or extract results. (2) Parameters lack descriptions in many cases (e.g., 'edits' array in readmd_pdf_apply_edit has no detail on structure). (3) No error handling guidance, tools return success/failure with no recovery hints. (4) Tool names are inconsistent in verb usage (readmd_* prefix adds noise; 'readmd_ai_assistant' is vague). (5) No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels. (6) Confirmation parameters exist but no guidance on how LLMs should handle them. Baseline: 549 production tools average 18-char names and 194-char descriptions; this server's descriptions are adequate length but lack actionable detail for LLM selection.
使用 ReadMD 共享 Skill Registry 执行兼容的文档 AI 流程
将本地 Word (.docx), PDF (.pdf), PPT (.pptx), Excel (.xlsx), LaTeX (.tex), HTML, TXT 等转为高质量 Markdown
将 Markdown 编译导出为排版级 PDF、Word (.docx)、HTML 或学术 LaTeX
将 Markdown 文档导出为 EPUB 电子书格式
将 Markdown 幻灯片编译导出为 HTML 演示文稿或 PDF
自动诊断并修复 Markdown 格式错误（公式断裂、表格错位、代码块、中英文标点与空格规范）
从 Markdown 文档生成目录（Table of Contents）
No output schemas documented for any tool. LLMs cannot infer result structure, forcing them to guess at field names and types. This breaks tool chaining and increases hallucination risk.
Parameter descriptions are minimal or absent. 'edits' in readmd_pdf_apply_edit is an array with no schema detail. 'parameters' in readmd_ai_assistant is an object with no field guidance. LLMs cannot construct valid payloads.
No error handling guidance. Tools declare risk levels (WRITE, IRREVERSIBLE) but provide no recovery hints. If a tool fails, LLMs have no guidance on retry strategy, user escalation, or compensation actions.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | <=2025-11-25 | v2 |
获取指定 Skill 的完整 Prompt 模板和说明
LaTeX 论文/公式源码精确转为标准 Markdown
将 LaTeX 公式转为 Word 原生 Office Math (OMML) / MathML XML
列出所有可用的 ReadMD Skills（AI 工作流模板）
Markdown 文档编译为标准学术 LaTeX 源码
本地图片或扫描件 PDF 通过 WinRT / macOS Vision / Tesseract 离线 OCR 识别为 Markdown
自动解析 BibTeX 文献数据库并生成学术引用映射
对 PDF 文档应用编辑操作（文本替换、删除、插入等）
回滚 PDF 编辑操作到之前的版本
处理 Markdown 文档中的导入指令（@import 语法），合并外部文件内容
执行 Markdown 代码块（Python、JavaScript 等）并返回输出
抓取网页 URL 并深度抽取清洗为高可读 Markdown
Tool name 'readmd_ai_assistant' is vague and does not follow verb_noun pattern. What does it do? Does it chat, generate, execute? The description mentions 'Skill Registry' but does not clarify the tool's actual function.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk classifications. Tools marked WRITE or IRREVERSIBLE should declare this in schema annotations so agents can reason about safety.