Already have configs?

Most people don’t start clean — you arrive with servers and plugins already configured in Claude or OpenCode. Bring them under management instead of retyping them.

1. See what’s on disk vs. what agentsync would write.

   agentsync status

2. Import native config into your canonical source.

   agentsync import claude --dry-run       # preview what a full import would write
   agentsync import claude                 # the agent's full native config

3. Apply to fan it out to your other agents.

   agentsync apply

The import selector

Import uses a selector grammar: <agent>[:<component>[:<name>]]. Drop parts from the right to widen the scope.

Selector	Imports
claude	the agent’s full native config (MCP, skills, subagents, commands, hooks, LSP, memory)
claude:mcp	every MCP server
claude:mcp:github	a single MCP server
opencode:mcp:linear	a single server from a different agent

agentsync import claude
agentsync import claude:mcp
agentsync import claude:mcp:github
agentsync import opencode:mcp:linear

A bulk import that finds nothing for a component reports it and exits cleanly rather than erroring. Add --dry-run to list the source files an import would write without touching ~/.agentsync/.

The first apply on a populated machine

On a machine that already has native config, the first apply sees files it didn’t write and treats them as a foreign-collision: it backs each one up to ~/.agentsync/.state/backups/<timestamp>/ before writing. Nothing is lost. Preview which files will be backed up with agentsync apply --dry-run first.

Why import, then apply?

Import pulls a native edit into your canonical source — re-referencing any resolved secret back to ${secret:…} form on the way in, so credentials never get committed. Once it’s in ~/.agentsync/, a normal apply fans it out to every other enabled agent.

The daily loop Now that your config is under management, here's the rhythm of day-to-day use.