Loading ToolBench
A remote MCP server providing calculator tools without authentication
This server has significant definition quality gaps. Both tools lack parameter descriptions entirely, which violates a critical MCP pattern requirement. The 'add' tool description is trivial (19 chars), falling below the actionable threshold. While basic schemas are present via Zod validation, they lack the descriptive context LLMs need for proper tool selection and parameter inference. The 'calculate' tool is slightly better with a 45-char description, but still lacks depth about when to use it vs 'add', what the enum values mean, and what error conditions might occur.
Simple addition tool
Calculator tool with multiple operations
All input parameters lack descriptions. The Zod schema defines types (number, enum) but provides no natural-language guidance for LLMs about what each parameter controls, valid ranges, or example usage.
'add' tool description is 19 characters ('Simple addition tool'), falling below the 20-character minimum actionable threshold. This brevity provides no context for tool selection, error handling, or when to prefer 'add' over 'calculate'.
Parameter 'operation' in 'calculate' uses an enum ['add', 'subtract', 'multiply', 'divide'] but provides no description of what each operation does, what the return type is, or error handling for division by zero (which IS handled internally but not documented).
No output schema documentation. LLMs cannot reason about what fields to expect from responses (e.g., is the result a number or string? Does it include metadata?). The code returns text content, but this is not documented.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 44 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 25 | - | v1 |
Error handling for division by zero returns a text message, but there is no schema-level error classification (retryable vs user-fixable vs fatal) and no guidance for LLMs on how to handle the error response.
No tool annotations present (readOnlyHint, destructiveHint, idempotentHint). While both tools are read-only and safe, this is not formally declared to clients.