A weather intelligence MCP server that provides global weather forecasts, alerts, air quality, UV index, and marine forecasts using Open-Meteo and NWS APIs, with support for multiple LLM providers (Anthropic, OpenAI, Gemini, Vertex AI)
The server provides 15 weather and location tools with generally complete schemas and descriptions. Most tools have clear action verbs (get_, geocode_) and non-empty descriptions. However, several quality issues reduce the overall score: (1) Descriptions lack detail on when/why to use tools and dependencies between them. (2) Some parameters lack descriptions (e.g., 'date' in get_sunrise_sunset has a description but lacks format constraints in schema). (3) No output schemas are documented, LLMs cannot infer what fields to expect from responses. (4) Error handling is minimal, no guidance on retries, timeouts, or recovery paths. (5) Tool composition lacks explicit chaining hints (e.g., geocode_location produces lat/lon; other tools consume them, but this dependency is not documented in descriptions). (6) No security or permission checks visible. The HTTP transport is good; tool naming is solid. Baseline comparison: avg param count is ~2.2 (within p10-p90 range); avg description length ~150 chars (reasonable but could be richer with context).
Compare current weather between two locations.
Convert a place name, address, or location to latitude/longitude coordinates. Use this first when you have a location name but need coordinates for other tools.
Get active tropical storms and hurricanes in the US.
Get air quality data for any location worldwide.
Get weather alerts for a US state.
Get current weather conditions from nearest observation station (US only).
Get weather forecast for a US location.
No output schemas documented. LLMs cannot infer what fields responses contain (e.g., does get_forecast return 'high_temp' or 'temperature_max'?). This forces agents to guess field names and breaks downstream tool chaining.
Tool compositions are implicit. geocode_location produces latitude/longitude that other tools consume, but descriptions lack chaining hints like 'Call geocode_location first if you only have a place name.' This wastes tokens on agent trial-and-error.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 62 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 55 | - | v1 |
Get 7-day weather forecast for any location worldwide. Returns daily high/low temperatures, precipitation, wind speed, and weather conditions.
Get hourly weather forecast for any location worldwide (next 24 hours).
Get hourly weather forecast for a US location (next 24 hours).
Get marine/ocean forecast for coastal locations worldwide.
Get nearby radar stations for a US location.
Get sunrise and sunset times for a location.
Get UV index for any location worldwide.
Get comprehensive weather summary including current conditions, forecast, and air quality.
Descriptions lack context on WHEN to use tools. get_forecast and get_global_forecast both fetch weather forecasts but differ by region (US vs worldwide). Descriptions do not clarify this trade-off, forcing LLMs to guess which to call.
No error handling or recovery guidance. If geocode_location returns no results, the description does not guide the LLM on next steps (try broader query? offer alternatives?). No timeout handling documented for external API calls.
Parameter descriptions are sparse. get_alerts accepts a 'state' string but does not validate format ('CA' vs 'california' vs 'ca'?). The description does not specify case sensitivity, format constraints, or error behavior.
No permission checks or audit trails visible. Tools make external API calls but do not declare what permissions they require (read:weather, etc.) or log access for compliance.
Descriptions contain example values (e.g., 'e.g. Paris, France', 'e.g. CA, NY'). LLMs tend to reuse these literally instead of adapting to context. Use enums or format constraints instead.