Loading ToolBench
MCP server for retrieving knowledge from Ragflow RAG system
Single tool with minimal definition quality. Tool name 'retrieve_knowledge' lacks action verb clarity (should be 'search_knowledge' or 'query_knowledge'). Description is adequate (67 chars) but generic. Input schema present but severely incomplete: only 'query' parameter with basic string type, no constraints, no format guidance. No output schema documented, tool returns TextContent but LLM cannot predict structure. No error handling guidance. No parameter validation rules. Fails multiple critical checks from pattern:tool and pattern:tool-description.
Retrieve knowledge related to the query from Ragflow
Tool name 'retrieve_knowledge' is a noun phrase, not a verb_noun action. Should be 'search_knowledge', 'query_knowledge', or 'retrieve_from_knowledge_base' to signal intent clearly to LLM.
Input schema lacks constraints and format guidance. 'query' parameter has no min/max length, no pattern, no examples of valid queries. LLM cannot validate input before calling.
No output schema documented. Tool returns TextContent with formatted knowledge, but LLM cannot predict structure (fields, format, pagination). Breaks pattern:tool and pattern:response-shaper.
Error handling returns raw exception messages ('请求 RAG 服务时出错: {exc}') without recovery guidance. LLM cannot determine if error is retryable or user-fixable.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | <=2025-11-25 | v2 |
Configuration (API key, dataset ID, thresholds) hardcoded in environment variables with no validation. If RAGFLOW_URL or RAGFLOW_API_KEY are missing, tool fails silently at runtime.