MCP servers
Model Context Protocol servers add tools to the agent. Managing them, committing a set for a whole team, and publishing Kaeonix's own tools to something else.
Adding a server
An MCP server contributes tools to the agent's registry, alongside the core tools, for the whole session. Manage them from the CLI:
On first run Kaeonix auto-imports existing Claude Code and OpenCode MCP configuration, so migrating from either takes no manual re-entry.
mcp listShow configured servers and whether each connects.mcp addAdd a server interactively.mcp remove <name>Drop a server from this config layer.mcp login | logoutAuthenticate against a server that requires OAuth.mcp servePublish Kaeonix's own tools over MCP.Sharing a set with the team
Commit .kaeonix/mcp.json and everyone working in the repository picks up the same servers. This is the right layer for anything project-shaped — an issue tracker, a schema browser, an internal search index:
Personal servers belong in ~/.kaeonix/ instead, so they follow you across repositories without landing in someone else's checkout. See Configuration for how the layers merge.
Serving Kaeonix's own tools
mcp serve exposes Kaeonix's tools as an MCP server, so another agent can drive them. Pair it with --read-only to publish only the non-mutating ones — no write, edit, bash, or network:
Trusting a server
Every connected server also costs context: its tool schemas are sent on every turn whether or not the agent calls it. Prune the ones you are not using — see Context window.