Loading ToolBench
A RAG (Retrieval Augmented Generation) chatbot system integrating local FAQ documents and official Datapizza-AI documentation via MCP, with multi-language support and memory management using Google Gemini 2.5 Flash
Scoring was not performed
No output schema documented. The 'ask' tool returns a string but provides no structured format guidance for downstream tool chaining or multi-step agent plans.
Parameter descriptions are minimal and in Italian, lacking actionable constraints. 'k' parameter has no range constraints (e.g., 1-100). 'score_threshold' has no guidance on valid range (0.0-1.0 assumed but not stated). LLMs cannot infer valid bounds.
Tool description is brief (Italian: 'Invia una domanda al chatbot...') and lacks WHEN to use it guidance. Does not explain that this is a RAG chatbot operating on FAQ data, limiting agent's ability to decide if this is the right tool.
No error handling guidance in tool definition. Code shows fallback messages and debug info collection, but tool schema does not document what errors can occur or how to recover (e.g., 'Qdrant connection failed' or 'No FAQ chunks retrieved above threshold').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 16 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 32 | - | v1 |
Unknown MCP transport. Source code shows custom Python framework ('datapizza') with no visible MCP protocol implementation. No evidence of SSE, HTTP, or STDIO MCP transport, appears to be a standalone chatbot, not an MCP server.
No pagination or result limiting documented. If the chatbot returns multiple chunks or a long FAQ response, there is no limit stated (baseline: cap at 20-50 items). Risk of blowing context windows on large FAQ sets.