Weather MCP Server by Meteosource

Weather MCP Server by Meteosource

Weather data for AI agents. Meteosource hosts the server at

https://mcp.meteosource.com/mcp

It speaks the Model Context Protocol over Streamable HTTP. Connect it to Claude Code, Claude Desktop, Codex, Antigravity or any other MCP client, send your Meteosource API key in a header, and the agent can look up forecasts, official weather warnings, air quality, twenty years of historical weather and the production of PV systems and wind farms.

There is nothing to install and nothing to host. The server keeps no key of its own: every request is served on the key it carries, within the plan that key is on.

Tools

The server tells the agent how the tools fit together, so a question like "will it rain in Brno tomorrow" ends up as a search_places call followed by get_forecast without you spelling that out.

Before you connect

You need a Meteosource API key. Sign up at meteosource.com (the free plan is enough to try the forecast tools) and copy the key from your account.

The key travels with every request in the Authorization header as Bearer <key>. Each setup below comes down to two values: the URL above and that header.

The rest the server works out itself. It asks Meteosource which plan the key is on and serves the tools that plan includes; a tool outside the plan answers with a sentence saying so. See Plans and limits.

Connect

Claude Code

One command:

claude mcp add --transport http meteosource https://mcp.meteosource.com/mcp \
  --header "Authorization: Bearer YOUR_METEOSOURCE_API_KEY"

This registers the server for the current project. Add --scope user to have it in every project. claude mcp list shows whether it connected; inside a session, /mcp lists the tools.

To share the setup with a team without sharing the key, commit a .mcp.json at the project root and let everyone keep their key in an environment variable:

{
  "mcpServers": {
    "meteosource": {
      "type": "http",
      "url": "https://mcp.meteosource.com/mcp",
      "headers": {
        "Authorization": "Bearer ${METEOSOURCE_API_KEY}"
      }
    }
  }
}

Claude Desktop and claude.ai

Connectors belong to your account, so one added on the web is also there in the desktop app and the other way round.

1. Open Customize > Connectors (in the desktop app: Settings > Connectors) and click Add custom connector.
2. Name it Meteosource; the URL is https://mcp.meteosource.com/mcp.
3. Under Authentication choose No sign-in.
4. Under Request headers add Authorization with the value Bearer YOUR_METEOSOURCE_API_KEY. Type the word Bearer and the space yourself: Claude sends the value exactly as you enter it.
5. Click Add. In a chat, open the + menu, choose Connectors and switch Meteosource on.

On Team and Enterprise plans an Owner adds the connector under Organization settings > Connectors and members connect to it from Customize > Connectors. The free plan allows one custom connector.

Codex

Codex CLI, the Codex app and the IDE extension read one file, ~/.codex/config.toml, so the server is configured once for all three. From a terminal:

codex mcp add meteosource --url https://mcp.meteosource.com/mcp \
  --bearer-token-env-var METEOSOURCE_API_KEY

and export METEOSOURCE_API_KEY in your shell profile. The section this writes to the file:

[mcp_servers.meteosource]
url = "https://mcp.meteosource.com/mcp"
bearer_token_env_var = "METEOSOURCE_API_KEY"

The Codex app has its own dialog (Settings > MCP servers > Add server, type Streamable HTTP), but no field for the key. If the app shows the server as failed, it did not get the variable: an app started from the Dock or the Start menu does not see what you export in a terminal. Put the key in the file instead:

[mcp_servers.meteosource]
url = "https://mcp.meteosource.com/mcp"
http_headers = { Authorization = "Bearer YOUR_METEOSOURCE_API_KEY" }

Codex gives a tool call 60 seconds by default. The first request for a year of archive at a location sends 365 requests to Meteosource and can take longer than that, so give this server more time:

tool_timeout_sec = 600

Restart Codex after editing the file (in the app, Restart on the MCP servers page). codex mcp list shows the result; /mcp in a session lists the connected servers.

Antigravity

The IDE and the CLI (agy) share one configuration file: ~/.gemini/config/mcp_config.json for all projects, or .agents/mcp_config.json inside a project. Add:

{
  "mcpServers": {
    "meteosource": {
      "serverUrl": "https://mcp.meteosource.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_METEOSOURCE_API_KEY"
      }
    }
  }
}

The key is written out literally; Antigravity does not expand environment variables in this file. If you use the project-level file, keep it out of version control.

In the IDE the same file is reachable from the agent panel: the … menu, MCP Servers, Manage MCP Servers, View raw config. In the CLI, /mcp opens the MCP manager, which shows the connection status and reloads the configuration.

Other clients

Any client that supports MCP over Streamable HTTP works with the same two values, the URL and the header. The server also accepts the key as X-API-Key: <key>, for clients that have a field for an API key but not for a whole header.

Try it

• What's the weather in Brno this weekend?
• Are there any weather warnings for Prague, Vienna and Munich right now?
• Compare last January in Berlin with the long-term normal, day by day.
• How many hours of wind above 10 m/s did 49.2, 16.6 see in 2025, month by month?
• Air quality in Kraków for the next two days, hour by hour.
• How much would a 10 kWp south-facing roof at 50.08, 14.43 have produced last week? (Renewables plan)

Plans and limits

One key is on one Meteosource plan, and the plan decides what the server serves. The weather tools are for the Free, Startup and Standard plans; what each plan includes, and how far back its archive reaches, is on the pricing page. The production tools need the Renewables plan, which in turn has no weather tools.

Days older than the plan's archive need Meteosource's historical add-on. Without it the API refuses such a day with "not allowed to access historical data this old"; the tool reports that for the location concerned and the rest of the call succeeds, so start inside the plan's range.

Per call: up to 20 locations; hourly archive and production up to 14 days, daily or monthly up to 366 days; at most 400 location-days (locations × days) in one call, so a long analysis is several calls. Each location, and each archive or production day, is one request against the key's quota, at the rate limit of its plan. The server keeps a cache, so repeating a call soon afterwards usually costs little or nothing. Cached data is cleared after a while, so a repeat after a longer gap counts against the quota again.

What the agent receives

Time series come back as tables, columns once and rows as arrays, with nested fields flattened (wind_speed, probability_storm, cloud_cover_total). That is four to seven times fewer tokens than the raw API: a week of hourly forecast is about 12 kB, a year of daily archive about 16 kB. The variables parameter trims it further when the agent only needs a few fields. Units default to the local convention of each location; units="metric" forces SI.

Your key and your data

• Your key is used only to call Meteosource on your behalf. It is never stored in full and never appears in a URL.
• Caches are kept per key. Nothing fetched for one key is ever served to another.
• The service records that a call happened, which tool it used and how it went, so the service can be operated.
• HTTPS only. A request over plain HTTP is refused, so a key cannot be sent in the clear by mistake.

Troubleshooting

The tools appear in the client without a key; only a tool call needs one. When a call fails, the agent gets a sentence it can show you.

Questions about your key or plan go to Meteosource support. For the server itself, open an issue in this repository.

About this repository

This repository holds the documentation of the hosted server and its entry for the MCP Registry, server.json. The server is listed there as com.meteosource/weather; clients and catalogs that read the registry pick it up from that entry. The server itself runs at Meteosource; there is nothing here to install.

License

MIT, see LICENSE.