Loading ToolBench
MCP server providing RAG (Retrieval-Augmented Generation) capabilities with document search and mathematical operations
Critical gaps across all dimensions. Only 1 of 4 tools has a description (search_documents). The 'add' tool lacks any description entirely. Parameter descriptions are minimal or absent for most tools. Schemas are present but inconsistent: some tools use dataclass wrappers (add, sqrt) while others use raw parameters (subtract). No output schemas documented. No error handling guidance. Tool naming is acceptable but lacks verb-noun consistency ('search_documents' is good; 'add', 'sqrt', 'subtract' are generic math operations that don't convey business intent). The server appears to be a proof-of-concept mixing RAG functionality with unrelated math utilities, suggesting poor composition and unclear purpose.
Search for relevant content from uploaded documents.
Square root of a number
Subtract two numbers
Missing tool descriptions: 'add' has no description at all. LLMs cannot determine when to select this tool.
Inconsistent parameter documentation. 'search_documents' has a query description; 'add', 'sqrt', 'subtract' have minimal or no parameter descriptions. LLMs cannot infer what 'a' and 'b' mean without explicit guidance.
No output schemas documented. Tools return raw integers/floats or lists without explaining structure. LLMs cannot plan downstream calls or extract fields reliably.
Poor tool composition. Math utilities (add, sqrt, subtract) are unrelated to the RAG domain (search_documents). This mismatch suggests the server lacks a clear purpose and will confuse LLMs about when to use each tool.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 37 | <=2025-11-25 | v2 |
No error handling or recovery guidance. search_documents catches exceptions and returns a generic error string; other tools have no error handling. LLMs receive no actionable guidance on what to do if a call fails.