Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Open-Meteo MCP Server What's the weather in Tokyo this weekend?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A small MCP (Model Context Protocol) server that gives an AI assistant like Claude the ability to look up real weather data — current conditions and forecasts — by wrapping the free Open-Meteo API. No API key required.
Built as a learning project to understand how MCP servers work end to end: defining tools, running them locally, and connecting them to a real AI client.
Once connected to an MCP-compatible client (like Claude), it exposes three tools:
Tool | What it does |
| Turns a place name ("London") into coordinates |
| Current temperature, wind, precipitation, etc. for a coordinate |
| Up to a 16-day forecast (daily and/or hourly) for a coordinate |
So a person can ask an assistant "what's the weather in Lisbon tomorrow?" and the assistant will call these tools itself to find out and answer — it isn't pre-programmed with weather data, it fetches it live.
Related MCP server: Weather MCP Server
src/server.ts — defines the three tools and how each one calls the
Open-Meteo API. This is the actual "brains" of the server.
src/index.ts — runs the server over stdio (standard input/output).
This is the transport used by local dev tools like the
MCP Inspector and by
desktop apps that launch the server as a subprocess.
src/http.ts — runs the same server over Streamable HTTP, so it
can be reached over the network — e.g. hosted somewhere, or reached from a
browser-based client. This is what makes it possible to demo without
installing anything locally.
Both entry points share the same tool definitions in server.ts, so there's
only one place where the actual logic lives.
npm install
npm run buildLocal/stdio mode (for tools that spawn the server as a subprocess):
npm startHTTP mode (for connecting a remote client, e.g. over a forwarded Codespaces port or a hosting provider):
node build/http.jsThis starts a plain HTTP server on port 3000 (configurable via PORT) with
a single MCP endpoint at POST /mcp, plus a GET /health check.
The MCP Inspector's CLI mode is the fastest way to poke at it directly from a terminal, no browser needed:
# See what tools are available
npx @modelcontextprotocol/inspector --cli node build/index.js --method tools/list
# Actually call one
npx @modelcontextprotocol/inspector --cli node build/index.js \
--method tools/call --tool-name geocode_location --tool-arg name=LondonClaude Desktop / Claude Code: point at build/index.js as a stdio MCP
server in the client's config.
Claude.ai custom connectors: point at the /mcp URL of a running HTTP
instance (requires a plan that allows custom connectors).
Anthropic API directly: pass the /mcp URL in the mcp_servers
parameter of a /v1/messages request — this lets any custom app or page
use the tools without needing a pre-built MCP client.
Open-Meteo has many more variables and endpoints than this project uses
(air quality, marine forecasts, historical/archive data, ensembles). To add
more, follow the pattern of an existing tool in src/server.ts — register
a new server.tool(...) block with its own input schema and Open-Meteo
request.
Data © Open-Meteo.com, CC BY 4.0.
Look up geographic coordinates (latitude/longitude) for a place name, e.g. 'London', 'Paris, TX', or 'Tokyo'. Use this before calling the weather tools if you only have a place name rather than coordinates.
Inputs: name (required), count, language.
Get the current weather conditions (temperature, wind, precipitation, etc.) for a specific latitude/longitude. Use geocode_location first if you only have a place name.
Inputs: latitude (required), longitude (required), timezone, wind_speed_unit, +1 more.
Get a multi-day hourly and/or daily weather forecast (up to 16 days) for a specific latitude/longitude. Use geocode_location first if you only have a place name. Defaults to a useful set of daily summary variables; pass include_hourly=true for hour-by-hour detail as well.
Inputs: latitude (required), longitude (required), timezone, forecast_days, +3 more.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 3 tools
Each tool has a clearly distinct purpose: geocoding converts place names to coordinates, while the other two retrieve weather data. The current weather and forecast tools are differentiated by time horizon and description.
All tool names follow a consistent snake_case verb_noun pattern: geocode_location, get_current_weather, get_weather_forecast. The naming style is uniform and predictable.
With 3 tools, the server is tightly scoped for its purpose: location resolution, current conditions, and forecast. Each tool earns its place and the count feels complete without redundancy.
The core weather lookup flow is fully covered: geocode a place name, get current conditions, or get a forecast. Minor gaps exist such as historical weather data or alerts, but these are not essential for the apparent primary use case.
Global weather via Open-Meteo: forecast, historical, marine, air quality, geocoding, elevation.
Current weather and forecast lookup as an agent tool.
Worldwide place search with coordinates (Open-Meteo) — paid per call (x402/credits), 1 tools
Get current weather for any city and create images from your prompts. Streamline planning, reports…