Agent Skill

The meteoswiss-ogd skill teaches an agent to read MeteoSwiss Open Data directly, using curl, awk and jq. There is no server, no Node.js and no API key. It is the lighter alternative to the MCP server for agents that have a shell, such as Claude Code or Cursor.

What it covers

Question	Skill	MCP server
Current weather	yes	yes
Forecasts	yes	yes
Station search	yes	yes
Pollen	yes	yes
Climate series (NBCN)	yes	yes
MeteoSwiss website search and fetch	no	yes
Structured, schema-validated output	no	yes

The skill ships five scripts (current-weather.sh, forecast.sh, pollen.sh, search-stations.sh, search-forecast-points.sh) and a reference document. For the cases it doesn’t cover, it points the agent to the MCP server.

Install

Skills CLI (works straight from the GitHub URL):

pnpx skills add https://github.com/eins78/meteoswiss-llm-tools.git#packages/meteoswiss-skills --global --agent claude-code --all

Claude Code plugin marketplace, from a local checkout (the marketplace manifest sits in the skills package, not at the repository root):

git clone https://github.com/eins78/meteoswiss-llm-tools.git

/plugin marketplace add ./meteoswiss-llm-tools/packages/meteoswiss-skills
/plugin install meteoswiss-skills@meteoswiss-marketplace

Manually: symlink packages/meteoswiss-skills/skills/meteoswiss-ogd into ~/.claude/skills/.

Which one should I use?

The Skill vs. MCP case study compares both: capability parity, effort, operations and context cost. Short version: choose the MCP server for structured output, broad coverage and many users; choose the skill when you want zero infrastructure and your agent has a shell.