MCP server providing weather alerts and forecasts from the National Weather Service (NWS) API
Two tools with basic verb-noun naming (get_alerts, get_forecast). Both have short descriptions (47 and 48 chars respectively) that meet minimum but lack LLM-optimized detail. Input schemas present with types and descriptions, but parameters lack constraints (enums, ranges, format specs). Output is unstructured text rather than structured JSON. No error recovery guidance. No pagination support despite potential for large result sets. Tool descriptions do not explain when to use each tool vs the other, or what data structure is returned.
Get weather alerts for a US state.
Get weather forecast for a location.
Descriptions too minimal (47-48 chars). Missing WHEN to use, WHAT is returned, and dependencies. LLMs cannot distinguish between get_alerts and get_forecast or understand output structure.
Input parameters lack constraints. 'state' parameter accepts any string; should be enum of valid US state codes (AL, AK, AZ, ...). 'latitude' and 'longitude' lack range constraints (-90 to 90, -180 to 180).
Output is unstructured text (string return type). No documented schema. LLMs cannot extract structured data (alert severity, forecast temperature) for downstream reasoning or tool chaining.
No error recovery guidance. make_nws_request() silently returns None on any exception; tools respond with generic 'Unable to fetch' messages. LLM has no actionable next step (retry? different params? different tool?).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
No pagination or result limits. get_forecast hardcodes limit to 5 periods, but get_alerts returns all features without limit. Large alert sets could exhaust context window.