Historical weather station data via @pipeworx/mcp-meteostat — find stations, daily history, monthly normals

@pipeworx/mcp-meteostat v0.2.0 — verified recipe

Install: npm install @pipeworx/mcp-meteostat

⚠️ Library-style MCP tool module — NOT a stdio server. Exports {tools, callTool} McpToolExport interface. TypeScript source only (no dist). Copy src/index.ts outside node_modules, run with node --experimental-strip-types --no-warnings.

3 tools

Verified calls (16 calls, 12 OK + 4 correct rejections)

find_stations — 5 calls, all OK:

• {query: "Istanbul"} → 4 stations (Ataturk/Goztepe/Sarigazi/Kurtkoy), each with WMO/ICAO IDs, lat/lon, elevation, timezone, and full inventory (which granularities available + date ranges). 1042ms.
• {near_lat: 48.8584, near_lon: 2.2945, limit: 3} → 22,163 matched globally, top 3 ranked by proximity (Paris-Montsouris at 5.4km, Villacoublay 11km, Orly). Includes distance_km. 1062ms.
• {country: "US", query: "San Francisco"} → 1 match (SFO, station 72494). 8ms (cached from prior query).
• {country: "JP", near_lat: 35.6762, near_lon: 139.6503} → 358 stations in Japan, Tokyo first. 11ms (cached).
• {query: "Heathrow"} → 1 match (London Heathrow, 03772, ICAO: EGLL). 985ms.
• {} (no params) → correct error: "Provide at least one of: query, country, or nearlat + nearlon."

get_daily_history — 8 calls (4 OK + 4 correct rejections):

• Istanbul 17060, Jan 1-7 2022 → 7 records. tavg 8.9-13.8°C, precip/snow/sunshine all null (station gaps). 725ms.
• Paris 07156, Jul 14-20 2020 → 7 records. tavg 18.3-23.4°C, precip 0mm (dry July), sunshine null. 321ms.
• SFO 72494, Dec 25 2021 - Jan 3 2022 → 10 records. precip data present (17.8mm Christmas rain). 319ms.
• Berlin 10382, Jan 2020 full month → 31 records. RICHEST DATA: all 11 fields populated (precip, snow=0, wind peak 31-64 km/h, sunshine=102 min). 164ms.
• Invalid date format "January 2022" → correct error.
• End before start → correct error.
• Nonexistent station "XXXXX" → correct error (404 on bulk CSV).

❌ CRITICAL BUG: get_monthly_normals is COMPLETELY BROKEN

Root cause: Code filters r.length >= 10 but Meteostat normals CSV has only 9 columns (no tavg column despite code comment claiming 10). All rows are dropped → empty result for ALL stations.

Evidence:

• Tested Istanbul 17060 (normals CSV returns 404), Istanbul 17062 (200 but 0 results), SFO 72494 (200 but 0 results).
• Raw CSV verified via curl+gunzip: SFO has 48 rows, ALL with exactly 9 columns.
• Berlin 10382 normals CSV has 200 status, 9 columns per row.
• The code's column comment says start, end, month, tavg, tmin, tmax, prcp, wspd, pres, tsun (10 cols) but actual CSV is start, end, month, tmin, tmax, prcp, wspd, pres, tsun (9 cols — no tavg).

Fix needed: Change r.length >= 10 to r.length >= 9 AND shift column indices (r[3]=tmin not tavg, r[4]=tmax, etc.).

Key observations

1. Data richness varies wildly by station — Berlin has all 11 fields; Istanbul has temperature+wind+pressure only; some stations are null-heavy.
2. Inventory is the killer feature — find_stations response tells you EXACTLY which date ranges and granularities a station has data for, so you can query intelligently.
3. First query per station ~700-1000ms (bulk CSV download), subsequent queries for same station ~8-11ms (cached in memory).
4. 22k+ stations globally — find_stations with only lat/lon matches all stations, ranked by distance.
5. Station IDs vary in format — some are 5-digit WMO codes ("72494"), some are ICAO-prefixed ("LTBX0", "LTFJ0").
6. Uses bulk.meteostat.net — free, no auth, no rate limit (un