Loading ToolBench
MCP server for 得到大脑（原Get笔记）web version. Provides access to notes list, note details with AI summaries, original transcripts, and audio attachments. Handles login session management via persistent browser profile.
Server has 6 tools with basic descriptions in Chinese. All tools have input schemas with type definitions, but descriptions lack depth and LLM-optimization guidance. Parameter descriptions are minimal (1-2 sentences). No output schemas documented. Error handling exists (NotLoggedInError, AuthExpiredError, GetnoteError) but recovery guidance is absent. Tool names follow verb_noun pattern (getnote_*) but are generic and don't clearly distinguish operations. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Composition is reasonable, each tool has single responsibility, but parameter naming could be more explicit (e.g., 'note_id' is good, but 'mode' enum values lack context).
获取笔记的原始音频 OSS 直链；传 dest 可直接下载到本地。
弹出浏览器手动登录得到大脑（原Get笔记），登录成功后自动保存会话。首次使用或登录彻底失效时调用；会阻塞直到登录完成或超时。
获取笔记详情（AI 智能总结正文、附件信息）。mode=json 返回原始结构。
获取笔记列表（含 AI 总结）。all=true 全量翻页，note_type 可按类型过滤（如 local_audio）。
查看得到大脑（原Get笔记）登录态与账号信息。
读取录音/会议笔记的原始逐字稿（未润色，带毫秒时间戳）。mode=json 返回 sentence_list。
No output schemas documented. Tools return complex objects (notes with attachments, transcripts with sentence_list, etc.) but LLM cannot predict structure. Forces agent to guess field names and types.
Descriptions are minimal (1-2 sentences, mostly in Chinese) and lack LLM-optimization guidance. No explanation of WHEN to use each tool vs. similar ones. 'mode' parameter in getnote_note and getnote_transcript lacks context for enum values (text vs json).
No tool annotations despite clear risk levels. getnote_login and getnote_audio are WRITE operations; getnote_status, getnote_notes, getnote_note, getnote_transcript are READ_ONLY. Annotations (destructiveHint, idempotentHint, readOnlyHint) would guide agent planning.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | <=2025-11-25 | v2 |
Error handling exists (NotLoggedInError, AuthExpiredError) but tool descriptions do not guide recovery. No mention of 'call getnote_login first' or 'retry with exponential backoff'. Agents cannot self-correct.
Parameter descriptions are sparse. 'timeout' in getnote_login lacks range (is 300 seconds reasonable? max?). 'limit' in getnote_notes has no bounds. 'note_type' enum values (local_audio, link, img_text, plain_text) are not formally constrained in schema.