Loading ToolBench
An MCP server deployed on Cloudflare Workers that provides OAuth2 authorization handling and exposes MCP tools via HTTP/SSE transport. Demonstrates remote MCP server deployment with OAuth2 provider integration.
Single tool 'add' with minimal but present description and schema. Tool name is clear and action-oriented. Schema uses Zod validation (a.number(), b.number()) which translates to proper JSON Schema types. Description is generic and lacks context about when/why to use it. No output schema documentation. Parameters have descriptions but are minimal (3-4 words each). No error handling guidance. Overall adequate but sparse, typical of a demo/template server.
Tool description lacks context and selection criteria. 'Adds two numbers together' does not explain when to use this tool vs. alternatives, what the output format is, or whether this is idempotent.
Output schema not documented. Response structure is inferred from code ({content: [{type: 'text', text: string}]}) but not explicitly described in tool definition or docstring. LLM cannot plan downstream behavior without knowing output shape.
Parameter descriptions are minimal (e.g., 'First number to add'). No mention of constraints: Do negative numbers work? Floats or integers only? What is the valid numeric range? Unbounded numbers may cause overflow or precision issues.
No error handling or recovery guidance. If inputs are invalid types (e.g., non-numeric), the tool will fail silently with no actionable error message for the LLM to self-correct.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 5 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 51 | - | v1 |