au-weather-mcp

mcp-name: io.ausdata/au-weather-mcp

Ask Claude about Australian weather and air quality and get real, current numbers — not "I don't have access to that data." This MCP server gives Claude (and other MCP clients like Cursor) live access to Australian weather + air-quality data via Open-Meteo, which aggregates Bureau of Meteorology observations under licence. 45 curated locations (every state capital + every regional centre over ~25k), postcode and place-name lookup, current observations, 16-day forecasts, 80+ years of historical data, and multi-location comparison.

Hosted access? For cross-source queries, webhooks, an always-on REST API, and a uniform response envelope across all 9 sources, see ausdata.io — free tier available (500 calls/mo, no card).

Companion to abs-mcp (ABS macro stats), rba-mcp (Reserve Bank), and ato-mcp (tax + charity register) — together the four cover Australia's most-asked public data.

What you can ask

Once installed, your LLM can answer questions like:

Every response carries a CC-BY 4.0 attribution string and a direct Open-Meteo URL the agent can cite back to the user.

Why Open-Meteo (not BOM directly)

BOM publishes their own JSON/XML endpoints, but they actively 403 non-browser User-Agents and have no documented commercial-use path below their ~$5k/yr Registered User Service. Open-Meteo:

• Aggregates BOM data under their existing licensing arrangements with national meteorological services
• Free tier is explicit and generous; commercial use is $30/mo with public terms
• Returns clean, versioned, schema-stable JSON with units alongside every value
• Covers historical data back to 1940 via their archive endpoint
• No API key, no User-Agent gymnastics

We attribute both Open-Meteo and BOM in every response.

Install

# After publish:
uvx au-weather-mcp

# Local dev:
uv pip install -e .

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "weather": {
      "command": "uvx",
      "args": ["--upgrade", "au-weather-mcp"]
    }
  }
}

The --upgrade flag makes uvx re-check PyPI on each Claude Desktop launch, so bug fixes propagate without manual cache refresh. Costs ~100ms at startup.

For a local checkout (before PyPI publish):

{
  "mcpServers": {
    "weather": {
      "command": "uv",
      "args": ["run", "--directory", "/absolute/path/to/au-weather-mcp", "au-weather-mcp"]
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "weather": {
      "command": "uvx",
      "args": ["--upgrade", "au-weather-mcp"]
    }
  }
}

Tools

Accepts almost any input shape

The location parameter on every tool resolves six different input shapes — agents and users don't need to know the curated key format:

Every response includes a location_resolution field with one of curated, state_alias, raw_coordinates, geocoded, or fuzzy_curated — so the agent (and the user) can see HOW the input was interpreted.

Curated locations

45 curated locations covering all 8 state/territory capitals plus 37 major regional centres (every AU population centre over ~25k). Anything outside the curated set still works via the place-name geocoder or postcode lookup.

Coordinates are anchored to the canonical BOM observation point for each city (e.g. Sydney = Observatory Hill, Melbourne = Olympic Park) so cross-checking against BOM's official observations is straightforward. See src/au_weather_mcp/data/curated/locations.yaml for the full registry.

Worked examples

"What's the weather in Sydney right now?"

latest(location="sydney")

Returns:

{
  "location_id": "sydney",
  "location_name": "Sydney",
  "state": "NSW",
  "latitude": -33.8607,
  "longitude": 151.205,
  "timezone": "Australia/Sydney",
  "period": {"start": "2026-05-12T11:30", "end": "2026-05-12T11:30"},
  "current": {
    "time": "2026-05-12T11:30",
    "temperature_c": 19.7,
    "apparent_temperature_c": 18.1,
    "relative_humidity_pct": 67,
    "precipitation_mm": 0.0,
    "cloud_cover_pct": 43,
    "pressure_msl_hpa": 1034.5,
    "wind_speed_kmh": 18.4,
    "wind_direction_deg": 149,
    "wind_gusts_kmh": 43.2,
    "weather_code": 1,
    "weather_description": "Mainly clear"
  },
  "source": "Open-Meteo (aggregates Bureau of Meteorology data under licence)",
  "attribution": "Weather data by Open-Meteo.com (https://open-meteo.com), licensed under CC BY 4.0...",
  "source_url": "https://api.open-meteo.com/v1/forecast?latitude=-33.8607&...",
  "server_version": "<package version, e.g. 0.3.3>",
  "location_resolution": "curated",
  "location_input": "sydney"
}

"How was Sydney in January 2020?"

get_weather(
  location="sydney",
  start_date="2020-01-01",
  end_date="2020-01-31",
  granularity="daily"
)

Returns 31 DailyAggregate rows with temperature_max_c, temperature_min_c, precipitation_sum_mm, and weather descriptions per day.

"7-day Melbourne forecast, hourly detail"

get_weather(
  location="melbourne",
  start_date="2026-05-12",
  end_date="2026-05-19",
  granularity="hourly"
)

Returns 168 hourly rows.

Date formats

start_date and end_date must be YYYY-MM-DD. The server validates both the regex and the semantic date — 2024-13-40 raises a clean ValueError rather than silently failing.

Trust contract

Every response carries: