I used a weather MCP server to plan a Menorca trip with Claude

I was planning a trip to Menorca with friends, kids included, and autumn in the Balearics can be anything from beach weather to a wild north wind. Normally I’d open three weather apps and guess. But I work at Meteosource, and we recently launched our MCP server, so I tried it on my own trip: Claude, connected with nothing but a free API key. Below are the real requests and responses, no demo setup.

You need a free API key from meteosource.com. Then add the hosted MCP server to your AI tool.

claude mcp add --transport http meteosource https://mcp.meteosource.com/mcp \ --header "Authorization: Bearer YOUR_API_KEY" Enter fullscreen mode Exit fullscreen mode Claude (web or desktop): Customize > Connectors > Add custom connector, paste https://mcp.meteosource.com/mcp, set Authentication to No sign-in and add the request header Authorization: Bearer YOUR_API_KEY. On Team and Enterprise plans, an organization owner adds it under Organization settings > Connectors.

Nothing to install. The key goes in a header, never into the chat.

I am planning a trip to Menorca from Saturday for one week. Give me trip and exploring tips according to the weather.

Claude picked the get_forecast tool and asked for two points in one call, one on the east of the island and one on the west, to see if the weather differs across Menorca:

{ "locations": [ { "lat": 39.889, "lon": 4.265, "label": "Mahón (east)" }, { "lat": 40.001, "lon": 3.84, "label": "Ciutadella (west)" } ], "sections": "daily", "days": 7, "units": "metric", "variables": [ "summary", "temperature", "wind", "precipitation", "probability", "cloud_cover" ] } Enter fullscreen mode Exit fullscreen mode Note the variables filter: the model asked only for what it needed for trip planning.

Then I told it which village we were staying in, on the south coast.

This time it requested a full week of hourly data for that exact spot. Here is one day of the response:

{ "columns": [ "day", "weather", "summary", "predictability", "temperature", "wind_speed", "wind_gusts", "wind_dir", "wind_angle", "cloud_cover_total", "cloud_cover_low", "cloud_cover_middle", "cloud_cover_high", "precipitation_total", "precipitation_type" ], "rows": [ [ "YYYY-MM-DD", "psbl_rain", "Light rain changing to cloudy in the afternoon. Temperature 17/20 °C.", 3, 18.2, 4.5, 8.3, "NE", 36, 90, 57, 32, 77, 2.5, "rain" ] ] } Enter fullscreen mode Exit fullscreen mode Time series come back as tables: column names once, then plain arrays. That's why 168 hours of hourly data fits comfortably in the model's context. The same data as raw API JSON would repeat every key on every hour.

From the hourly data, Claude built a plan for our first long weekend:

The part I liked most: it didn't just read me temperatures. It connected the wind direction with geography. A north wind means rough water on the north coast and calm coves in the south, so it sent us south on the windy day.

The real value showed up in the follow-ups, because the weather data stayed in context:

The free key was enough for all of the above, but it has limits worth knowing:

With "units": "metric", wind speed comes back in m/s, not km/h. Day 3's 8.8 m/s north wind is about 32 km/h, which is very different from "a light 9 km/h breeze". If you build anything on top of this, convert explicitly or ask the model to show km/h.

Setup guides for Claude, Codex, Cursor, Antigravity and other tools are at mcp.meteosource.com.

If you try it for your own trip, tell me in the comments where you're going.