Aviation Weather MCP — METAR, TAF, PIREPs, AIRMET/SIGMET, station info
Aviation Weather MCP has 8 well-defined tools with complete input schemas and descriptions. All tools are READ_ONLY and domain-appropriate. However, several critical gaps reduce quality: (1) tool names lack action verbs (metar, taf, pirep instead of get_metar, fetch_taf, search_pirep); (2) parameter descriptions are inconsistent, some lack detail on constraints and formats; (3) no output schemas documented; (4) missing enum constraints on format and region parameters; (5) airmet tool has minimal parameter description ('format' with no explanation). The codebase shows solid implementation (proper error handling for PIREP bbox defaults, client-side filtering for SIGMET/AIRMET), but the tool interface itself needs refinement for LLM clarity.
Fetch all currently active AIRMETs (moderate turbulence, icing, IFR/mountain-obscuration advisories for general aviation) from the US Aviation Weather Center; no inputs required.
Fetch Graphical AIRMETs from the US Aviation Weather Center; optionally filter by type (turb, ice, sfc_winds, lifr, ifr); returns GeoJSON-style hazard polygons.
METAR observations. ids = comma-sep ICAO codes (KSFO,EGLL,LFPG).
Pilot reports (PIREPs). Call with no args for recent US-wide reports (default CONUS box), or pass a station `id` for reports within `distance_nm` of it, or a custom `bbox`.
Fetch all currently active SIGMETs (significant en-route weather hazards: convection, turbulence, icing, volcanic ash) from the US Aviation Weather Center; no inputs required.
Aerodrome metadata — lat/lon, elevation, country.
Tool names lack action verbs. Names like 'metar', 'taf', 'pirep' are domain nouns, not verbs. LLMs infer intent from verb-first names (get_metar, fetch_taf, search_pirep). Current names force LLMs to read descriptions to understand what action occurs.
airmet tool has minimal parameter description. The 'format' parameter lacks explanation (just type string, no description of valid values or default). This violates the requirement that every parameter must have a non-empty description explaining what it controls.
No output schemas documented. Tools return JSON/raw text but LLMs have no schema for expected fields. This prevents agents from planning downstream calls and extracting data reliably. E.g., does metar return { icao, time, wind, visibility } or a different structure?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
Terminal Aerodrome Forecast (TAF) for one or more ICAO airport codes (e.g. KSFO,EGLL); returns forecast periods covering wind, visibility, and cloud ceilings for the next 24–30 h.
Winds and temperatures aloft forecast from the US Aviation Weather Center; specify US region (us/bos/mia/chi/dfw/slc/sfo/alaska) and level (low/high); returns an ASCII table or JSON.
Missing enum constraints on format and region parameters. 'format' accepts 'json|raw|xml|html' but is declared as free-form string. 'region' accepts 'us|bos|mia|chi|dfw|slc|sfo|alaska' but no enum. LLMs will hallucinate invalid values without formal constraints.
Parameter descriptions lack format/range details. E.g., 'hours_before' says '1-24 (default 1)' but no explicit min/max in schema. 'distance_nm' says 'default 200' but no bounds. LLMs cannot validate inputs without formal constraints.