weather
# Weather Tool Calling with Cline and MCP
A small Python MCP server that lets an LLM fetch real weather data.
The user asks a weather question in Cline. The LLM decides whether to call a
weather tool, generates the required arguments, and sends the request to this
server. The server fetches data from the US National Weather Service (NWS),
then the LLM uses the result to write the final answer.
## How it works
```text
User question
-> Cline sends the question and tool definitions to the LLM
-> The LLM selects a weather tool and generates its arguments
-> Cline calls the Python MCP server
-> The server fetches real data from api.weather.gov
-> The LLM turns the tool result into a natural-language answer
```
## Tech stack
- Python 3.14
- MCP Python SDK
- Cline as the MCP client and LLM host
- OpenRouter with a DeepSeek model for the demo
- HTTPX
- US National Weather Service API
- uv
## Tools
### `get_forecast`
Gets the next five forecast periods for a US location.
```json
{
"latitude": 40.7128,
"longitude": -74.006
}
```
### `get_alerts`
Gets active weather alerts for a US state.
```json
{
"state": "NY"
}
```
## Setup
Requirements:
- Python 3.14
- [uv](https://docs.astral.sh/uv/)
- Visual Studio Code with Cline
- An API key for a tool-capable LLM configured in Cline
Install the project:
```bash
git clone <https://github.com/yang648557392/weather-mcp-tool-calling.git>
cd weather
uv sync
```
Add the MCP server to Cline's MCP settings. Replace the path with the absolute
path to this project:
```json
{
"mcpServers": {
"weather": {
"command": "uv",
"args": ["--directory", "/absolute/path/to/weather", "run", "weather.py"],
"disabled": false
}
}
}
```
Restart the MCP server in Cline. Cline should discover `get_forecast` and
`get_alerts` automatically.
If Cline cannot find `uv`, replace `"uv"` with the absolute path returned by:
```bash
which uv
```
## Usage
Ask Cline a question such as:
```text
What will the weather be like in New York tomorrow?
Are there any active weather alerts in California?
```
Cline will show the selected tool, its arguments, the tool result, and the
LLM's final response.
## Limitations
- The NWS API only supports locations covered by the United States weather
service.
- The LLM is responsible for converting a location name into coordinates.
- Cline is required as the LLM host and MCP client.
- Provider API keys are stored in Cline and must not be committed to this
repository.
## Author
Mingzhe Yang
TDQS
Scored across 2 tools
The two tools are clearly distinct: get_alerts retrieves weather warnings for a state, while get_forecast retrieves forecast data by coordinates. There is no overlap in their inputs or outcomes.
Both tools follow the same 'get_noun' pattern, with get_alerts and get_forecast. The naming is predictable and consistent.
With only two tools, the server is minimal and borderline scoped. While this could be fine for a specialized alerts/forecast service, it feels thin for a general weather service and does not reach the 3-15 tool sweet spot.
A weather service would typically include current conditions, hourly/daily details, or location-based lookup beyond forecast and alerts. The absence of these leaves significant gaps for users expecting general weather coverage.