Connect JMA Weather to Codex

Query JMA Weather directly from Codex. Arlet links the two through one MCP connection and takes care of API keys.

This server connects your AI client to the weather forecasts, precipitation forecasts and surface observations published on the Japan Meteorological Agency (JMA) website (www.jma.go.jp).

The forecast tools divide up the time axis, and each one names the next tool along, so you never hit a dead end:

  • 0-60 minutes: get_precipitation_nowcast — precipitation right now and in 5-minute steps for the next hour at an exact latitude/longitude, read from the 1 km High-resolution Precipitation Nowcast mesh. Optionally include recent observations to see whether rain is strengthening.
  • 1-15 hours: get_short_range_precipitation_forecast — hourly precipitation at the same point from the Short-range Precipitation Forecast, plus the latest analysed rainfall.
  • Up to 48 hours: get_distribution_forecast — 3-hourly precipitation, temperature and weather at an exact point from the distribution forecast mesh; and get_three_hourly_forecast — the same 48 hours for a forecast subdivision, with temperature as an actual number in °C plus weather and wind.
  • Days ahead: get_forecast — the 3-day prefectural forecast and the weekly forecast with climatological normals and the plain-text overview.

Also included:

  • search_area — resolve a Japanese place name (prefecture, region, or municipality; Japanese, kana, or English) to the JMA forecast area code. If nothing matches, every available forecast area is listed so you can pick one directly.
  • get_amedas_observation — the latest AMeDAS surface observations (temperature, precipitation, wind direction/speed, humidity, sunshine, snow depth in winter, etc.) for stations looked up by place name, station code, or nearest to a latitude/longitude.

Important notes:

  • No API key or account is required; the data is provided free of charge.
  • JMA publishes the mesh data as map tiles coloured by legend band, not as numbers, so the mesh tools return intensity as a band with a lower and upper bound (e.g. 1-5 mm/h, or temperature in 5 °C steps). When you need the temperature as an actual number and a forecast subdivision is precise enough, use get_three_hourly_forecast.
  • Outside JMA's composite radar area (remote islands such as Minamitorishima, or outside Japan) the mesh tiles are blank, so those locations always come back as no precipitation. Every mesh response reports how many frames could not be retrieved, so 'no value' is never confused with 'could not read'.
  • For weather warnings and advisories, landslide / inundation / flood risk (キキクル), thunder and tornado nowcasts, earthquakes and tsunami information, use the JMA Hazard server instead.
  • The data comes from JMA website content, not an officially guaranteed public API, so URLs and formats may change without notice.
  • Source attribution: data from the Japan Meteorological Agency website (気象庁ホームページ). Each tool response includes the attribution.
  • This server strictly relays forecasts and observations issued by JMA as-is and never generates its own forecasts (issuing original weather forecasts is restricted under Japan's Meteorological Service Act).
  • For life-threatening situations always check the JMA website or official broadcasts directly; this server is not a real-time alerting system.

How to connect JMA Weather to Codex

  1. 1

    In the Arlet dashboard, open Data tools, find JMA Weather and switch it on. No API key setup is needed.

  2. 2

    In your terminal, run codex mcp add arlet --url https://mcp.arlet.dev/mcp/ to add Arlet as an MCP server.

  3. 3

    Codex detects OAuth support and opens your browser. Sign in with Google and click Allow on the consent screen.

  4. 4

    Once connected, both the CLI and the IDE extension can call Arlet tools (the configuration is shared).

If the browser does not open, run codex mcp login arlet to redo the authentication.

Use JMA Weather with other AI clients