agentleFS
OlehDatsykOlehDatsyk/weather-mcp-server/.claude/skills/SKILL.md

weather-mcp-server

Skill1 starsChanged 17 days agoMITView on GitHub

Answers weather questions with live OpenWeatherMap data through the weather-mcp-server MCP tools - current conditions, 1-5 day forecasts, multi-city comparisons, air quality and official weather alerts. Use when the user asks "what's the weather in...", "will it rain this week in...", "compare the weather in X and Y", "what's the air quality in...", "are there any storm warnings for...", or wants trip-planning or what-to-wear advice based on the weather.

  • Reads credentials

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 6 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    This skill does not contain the server: If the tools listed below are not available in your session, follow "Setup check" before doing …

    line 13
  2. Distinctive:

    Use it when the user wants current or near-future weather facts for a named place:

    line 19
  3. Distinctive:

    Do not use these tools, and say plainly that the server cannot do it, when the user asks for:

    line 33
  4. Very specific:

    Look for tools named get_current_weather, get_weather_forecast, compare_cities, get_air_quality and get_weather_alerts.

    line 48
  5. Very specific:

    If none are available, tell the user the weather-mcp-server is not connected and point them to references/setup.md in this skill.

    line 52
  6. Very specific:

    If unsure whether an API key is set, read the resource weather://api/info.

    line 55
  7. Very specific:

    Identify the place: Use the city name the user gave.

    line 62
  8. Very specific:

    Choose units: Use the unit system the user asks for or clearly implies (°F for a US user -> imperial).

    line 66
  9. Distinctive:

    Check the result for an error: before using it (see "Error handling").

    line 71
  10. Distinctive:

    Always state the units (°C/°F/K, m/s or mph) and the city and country returned by the tool, so the user can spot a wrong match.

    line 115
  11. Distinctive:

    Never retry the same failing call more than once.

    line 153

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Weather MCP Server Skill

This skill tells you how to use the weather-mcp-server MCP server well. The server is a separate Python program that the user installs and runs locally. It exposes five tools, three resources and three prompts, all backed by the OpenWeatherMap API.

This skill does not contain the server. If the tools listed below are not available in your session, follow "Setup check" before doing anything else. Never invent weather data.

When to use this skill

Use it when the user wants current or near-future weather facts for a named place:

  • Current conditions: temperature, "feels like", humidity, wind, cloud cover, visibility, sunrise and sunset.
  • A forecast for the next 1-5 days in 3-hour steps.
  • A side-by-side comparison of 2-10 cities right now.
  • Current air quality (AQI 1-5 and pollutant levels).
  • Active official weather alerts (storms, floods, heat, etc.).
  • Advice built on the above: what to wear, which day is best for a trip, whether to pack an umbrella.

When NOT to use this skill

Do not use these tools, and say plainly that the server cannot do it, when the user asks for:

  • Historical weather or climate averages (e.g. "how hot was July 2019"). The server only returns current data and a forecast.
  • Forecasts beyond 5 days, hourly-only (1-hour) forecasts, or minute-by- minute rain.
  • Coordinates, postcodes or airport codes as input. Every tool takes a city name only.
  • Weather maps, radar images, marine, aviation or pollen data.
  • Air-quality forecasts or historical air quality (only current values).
  • Anything not about weather.

Setup check (do this first)

  1. Look for tools named get_current_weather, get_weather_forecast, compare_cities, get_air_quality and get_weather_alerts. Some clients add a prefix, for example mcp__weather__get_current_weather in Claude Code. Treat prefixed names as the same tools.
  2. If none are available, tell the user the weather-mcp-server is not connected and point them to references/setup.md in this skill. Do not guess the weather instead.
  3. If unsure whether an API key is set, read the resource weather://api/info. If api_key_configured is false, tell the user to set OPENWEATHER_API_KEY (see references/setup.md) before calling any tool.

Step-by-step workflow

  1. Identify the place. Use the city name the user gave. If the name is ambiguous (Springfield, Paris in Texas, London in Ontario), add an ISO country code: "Springfield,US", "London,GB". If you cannot tell which one they mean, ask once.
  2. Choose units. Use the unit system the user asks for or clearly implies (°F for a US user -> imperial). Otherwise leave the default (metric). Valid values: metric, imperial, standard (Kelvin).
  3. Pick the right tool using the table below. Use one call where one call is enough.
  4. Check the result for an error before using it (see "Error handling").
  5. Answer in plain language, using the output format below.

Tool selection

User wantsToolKey arguments
Weather now in one cityget_current_weathercity, optional units
Weather over the next daysget_weather_forecastcity, optional units, days 1-5 (default 3)
Several cities now, side by sidecompare_citiescities (list of 2-10 names), optional units
Air quality / pollutionget_air_qualitycity
Storm/flood/heat warningsget_weather_alertscity

Parameter rules:

  • days outside 1-5 is silently clamped by the server. If the user asks for 7 or 10 days, call with days: 5 and tell them 5 days is the limit.
  • compare_cities needs at least 2 and at most 10 cities. For one city use get_current_weather. For more than 10, split into several calls and say so.
  • get_air_quality and get_weather_alerts take no units argument.
  • For "trip" questions, call get_weather_forecast and get_weather_alerts for the destination.

Full schemas and every output field: references/tools.md.

Resources and prompts

Resources (read-only, no API call to OpenWeatherMap):

  • weather://cities/supported - 15 example cities. This is not a limit; any city OpenWeatherMap can geocode works.
  • weather://units - what each unit system measures in.
  • weather://api/info - provider, endpoints used, free-tier notes, and whether an API key is configured.

Prompts (the user may start these from their client, e.g. as slash commands): weather_summary(city), travel_recommendation(city, days=3), clothing_recommendation(city). If a prompt is used, follow its instructions; they call the same tools.

Expected output

  • Lead with the direct answer ("It's 14 °C and light rain in London").
  • Always state the units (°C/°F/K, m/s or mph) and the city and country returned by the tool, so the user can spot a wrong match.
  • Forecast and observation timestamps from the server are in UTC. Say so, or convert to the city's local time if you are sure of its offset.
  • For forecasts, summarise by day (high/low, main conditions, chance of rain from pop_percent), then highlight the best and worst periods. Do not paste all 3-hour entries unless asked.
  • For comparisons, give a short table and name the warmest, coldest and most humid city from the result fields.
  • For air quality, give the aqi_label (Good, Fair, Moderate, Poor, Very Poor) and mention PM2.5 and PM10 in µg/m³.
  • For alerts, list each alert's event, severity, issuing body and time window. If the list is empty, repeat the server's note accurately - an empty list may mean "no alerts" or "alerts not available on this API plan"; never claim "no warnings" when the note says alerts are unavailable.
  • End with the attribution line "Weather data © OpenWeather" (OpenWeather's licence asks for visible attribution).

Examples: see the examples/ folder.

Error handling and recovery

On failure a tool returns a JSON object with "error": true, an error_type and a message instead of weather data. Handle it like this:

error_typeWhat to do
city_not_foundAsk the user to check the spelling, or retry once with a country code ("City,CC").
missing_api_keyStop. Tell the user to set OPENWEATHER_API_KEY for the server and restart their client.
upstream_api_errorRead the message. HTTP 401: key invalid or not yet active (new keys can take up to a few hours). HTTP 429: rate limit - wait before retrying, do not loop. Other codes: report the message.
network_errorThe server already retried. Tell the user OpenWeatherMap could not be reached; suggest trying again later.
internal_error / service_errorReport the message and suggest restarting the server.

In compare_cities, one city failing does not fail the whole call: that entry has an error text and should be reported as "unavailable" while the other cities are still shown.

Never retry the same failing call more than once. Never fill gaps with estimated or remembered weather.

Security and privacy

  • Never ask the user to paste their API key into the chat, and never print, log or repeat a key. The key belongs in the server's .env file or in the client's MCP config env block on the user's own machine.
  • The only data sent to OpenWeatherMap is the city name (and coordinates derived from it). Do not add other personal details to the city argument.
  • City names are logged by the server in its local logs/ folder. Mention this if the user asks about privacy.
  • This skill gives no instructions to run shell commands. The optional scripts/verify_server.py is for the user to run themselves; it is read-only.

Quick examples

  • "What's the weather in Kyiv?" -> get_current_weather({"city": "Kyiv,UA"})
  • "Will it rain in Edinburgh this weekend?" -> get_weather_forecast({"city": "Edinburgh,GB", "days": 5}), then report only the weekend days.
  • "Is it warmer in Lisbon or Madrid?" -> compare_cities({"cities": ["Lisbon,PT", "Madrid,ES"]})
  • "Is the air safe to run in Delhi today?" -> get_air_quality({"city": "Delhi,IN"}) - report the AQI label; do not give medical advice beyond the AQI category.
  • "Any storm warnings for Miami?" -> get_weather_alerts({"city": "Miami,US"})

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.