Loading ToolBench
This is the official MCP TypeScript SDK, not a production MCP server. The three tools examined are example/test implementations in client code, not server tool definitions. Tool schemas are present but minimal. Descriptions are adequate (20-50 chars) but lack LLM-optimization guidance. The 'start-notification-stream' tool appears twice with slightly different configs, suggesting test/example duplication rather than production tooling. No evidence of error handling patterns, permission gates, or security hardening. The SDK itself is well-structured (Express, Zod validation, proper TypeScript), but the example tools do not meet production quality baselines.
Get weather information for a city
Starts sending periodic notifications
Starts sending periodic notifications for testing resumability
Duplicate tool definition: 'start-notification-stream' appears twice with different parameter defaults (count: 10 vs 50, interval: 1000ms vs 100ms). This violates single-responsibility and creates ambiguity for LLM tool selection.
Tool descriptions lack LLM-optimization guidance. 'Get weather information for a city' (34 chars) and 'Starts sending periodic notifications' (37 chars) are below the 50-200 char baseline for A+ tools. Missing WHEN to use, prerequisites, and return value hints.
No error handling patterns visible. Tools lack recovery guidance (e.g., 'If city not found, try search_cities()'), error classification (retryable vs fatal), or actionable error messages. Agents cannot self-correct on failure.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2025-06-18+ | v2 |
Parameter 'country' in get_weather lacks enum constraint. Description says 'Country code (e.g., US, UK)' but does not enforce valid ISO 3166-1 alpha-2 codes. LLMs will hallucinate invalid values like 'USA' or 'United States'.
Output schemas not documented. Tools return CallToolResult but no schema details visible for what fields agents should expect (e.g., does get_weather return temperature, humidity, forecast?). Agents cannot plan downstream calls.