Loading ToolBench
Multi-Agent AI system for Ziwei Doushu (紫微斗數) fortune-telling analysis with RAG knowledge base, web search, and multi-agent coordination. Provides MCP tools for generating and analyzing astrological charts.
This MCP server has significant quality gaps. Of 2 tools, both have schemas present but descriptions are minimal (Chinese, under 50 chars), parameter documentation is sparse, and there is no evidence of error handling, output schema documentation, or recovery guidance. The ziwei_chart tool description is 11 characters ('獲取紫微斗數命盤'), which violates the 10-1024 character baseline and provides no guidance on WHEN to use it or WHAT it returns. The search_ziwei_knowledge description (74 chars) is slightly better but still lacks specificity on output format or error recovery. Both tools lack documented output schemas, critical for guiding downstream agent planning. Input parameters have type definitions but descriptions are minimal or Chinese-only, making them harder for English-speaking LLMs to parse. No evidence of error classification, retryable vs. fatal categorization, or actionable error messages. The server appears to be an HTTP FastAPI backend but the actual MCP tool registration code is not fully visible in the provided source, only tool names and schemas are shown, suggesting tool definitions may be inferred rather than explicitly inspected.
搜索紫微斗數專業知識庫（RAG）。適合查詢星曜定義、宮位解釋、四化含義、命盤解讀規則等傳統命理知識。
獲取紫微斗數命盤
Tool descriptions are too short and lack actionable guidance. 'ziwei_chart' is only 11 characters (Chinese).
Output schemas are not documented. LLMs cannot infer what fields ziwei_chart returns (is it star positions, fortune summary, text analysis?). Without documented return types, downstream tool chaining is impossible.
Parameter descriptions are missing or language-mismatched. 'birth_hour' description is '出生時辰（子、丑、寅、卯、辰、巳、午、未、申、酉、戌、亥）' (Chinese zodiac hours), a non-English-fluent LLM will struggle. The enum constraint is present but description clarity is low.
No error handling or recovery guidance. Neither tool documents what happens on invalid input (e.g. invalid birth date Feb 31), network failures, or RAG knowledge base misses. No actionable error messages.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 45 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 25 | - | v1 |
Tool definitions appear inferred rather than explicitly visible in source. The provided source shows tool metadata but not explicit MCP tool registration code (e.g., no server.tool(name=..., description=...) calls visible). This limits confidence in definition completeness.
search_ziwei_knowledge lacks output schema. Does it return a list of knowledge chunks, a summary, raw text? How does the LLM parse results? Without documented structure, the agent cannot reason about what to extract.
Parameter 'query' in search_ziwei_knowledge has minimal description ('查詢關鍵字，例如...'). While it includes a Chinese example, English LLMs may misunderstand the domain context. Descriptions should be language-neutral or include English equivalents.