Skip to main content
The MCP server gives your assistant the tools. The skills teach it how to use them: call whoami first, never guess which organization a phrase meant, check what a report is called before running it, and quote the window with every number.
One command for Claude, Codex, Gemini CLI, opencode — anything that reads SKILL.md. Copying the files out of the repository does the same without the CLI.
The skills teach a model how to operate Glood, but they carry no tools of their own. Until you add the MCP server there is nothing for them to drive.

​
The two skills

They are split on purpose. A client loads a skill’s body only when its description fires, and most sessions ask for a named report and never compose a query — those sessions should not carry the vocabulary for one.

​
Claude Code gets both at once

The plugin carries the server and the skills, so neither step above applies:

​
claude.ai and ChatGPT need nothing installed

Neither surface loads SKILL.md files. The server sends both skill bodies itself when the connection opens, in the MCP instructions field, so the same guidance arrives with nothing installed. This is why there is no “install the skills” step for a connector — only for a CLI.

​
Keeping them current

Nothing in the skills is configuration. The organization list, the report catalogue, each agent’s instructions and every permission come from the server at call time — so a change you make in the dashboard reaches your next conversation with no reinstall, and nothing you installed can go stale in a way that matters.

​
Checking they loaded

Ask:
“What Glood organizations can you see?”
If the assistant does not recognise whoami, the server is connected but the skills are not loaded. Re-run npx skills add LoopClub/glood-agent-skills and start a new session.