Plugins

A plugin gives agents extra skills and, optionally, MCP servers. Manage plugins in Settings → Plugins. Each environment has its own plugins, installed on that environment's machine.

Add a plugin

Choose Add plugin, then either paste a git repository URL (with an optional branch or tag) or give the path of a folder on the environment's machine. Moonfold reads three layouts as they are:

• a Codex plugin (.codex-plugin/plugin.json, with an .mcp.json for its servers),
• a Claude Code plugin (.claude-plugin/plugin.json, with skills/ and .mcp.json),
• a plain folder of skills, each skill in its own folder with a SKILL.md.

A folder is copied when you add it. After you edit the original, choose Reload from folder. A repository plugin updates with Update from repository.

Turn a plugin on

Plugins start turned off, so nothing from a plugin reaches an agent until you choose to. When you turn on a plugin that has MCP servers, Moonfold lists the programs it lets agents start on that machine. Details shows the skills and servers at any time.

If an update changes a plugin's MCP servers, Moonfold turns the plugin off again so you can review them before you turn it back on.

Which agents use plugins

Enabled plugins reach new Codex and Claude Code turns. A skill is available to the agent like its own skills; a plugin's MCP tools ask for approval under the thread's permission mode. Other providers do not load plugins yet.

Turning a plugin on or off applies from the next turn. In a Codex thread that is already running, a plugin's MCP servers change when the thread's session next starts.

Claude Code starts a plugin's MCP servers in the project folder. A server written for Claude Code refers to its own files through ${CLAUDE_PLUGIN_ROOT}, which works. A server that expects to start inside the plugin's folder works with Codex but may not with Claude Code.