Share skills and MCP servers across Claude Code accounts
Every Claude Code configuration folder is a complete setup, so a second account starts without the skills, plugins and MCP servers you already added. You can add them again by hand, or share the ones you choose while keeping credentials and cloud connectors separate.
A guide from Switch, an independent Claude Desktop profile manager. Switch is in early access ahead of its public release.
Why a second account starts empty
Running a second account with its own CLAUDE_CONFIG_DIR keeps logins and history apart, which is the point. It also means MCP servers in .claude.json, your skills and your plugins live in the first folder only. Over time the two setups drift: a server you fixed in one is still broken in the other.
Add them in each account by hand
The direct approach is to repeat each setup step with the other folder active, for example running claude mcp add from a session started with that account’s CLAUDE_CONFIG_DIR, and copying skill folders into its skills directory. It works, but you have to remember every change for every account.
Avoid copying whole configuration folders: they contain that account’s session history and login state, which is exactly what you want to keep apart.
Share them with Switch
Switch shares supported local skills, commands, plugins and MCP server definitions across your profiles and your usual Claude Code setup on this Mac, the one the claude command uses in Terminal. You choose what to share and which profiles get it, so a database server can stay on Work only.
Sharing copies definitions, not secrets. Credentials aren’t copied, existing choices aren’t overridden, and each account still authorizes its own cloud connectors. How Switch handles Claude Code.
Decide what should stay separate
Not every tool belongs in every account. Keep servers that reach company systems in the work profile, and keep personal automations out of client accounts. Sharing a definition doesn’t grant another account access to a cloud service; that account still has to sign in to it, which is a useful check before work data crosses over.