Running Personal and Work Claude Code Accounts on One Machine

Run two Claude Code accounts (personal and work) on the same machine, side by side, without logging out each time.

How it works

Claude Code stores everything that makes an account "yours" — auth tokens, settings, history, projects, plugins, agents, hooks — inside a single config directory (default ~/.claude).

Setting the CLAUDE_CONFIG_DIR environment variable points Claude at a different directory. Each directory is a fully isolated profile.

Setup

Add these aliases inside the if status is-interactive block in ~/.config/fish/config.fish:

alias claude-personal "CLAUDE_CONFIG_DIR=$HOME/.claude-personal claude"
alias claude-work     "CLAUDE_CONFIG_DIR=$HOME/.claude-work claude"

Reload:

source ~/.config/fish/config.fish

Then run each alias once. Claude Code creates the config directory on first launch and prompts you to log in with the matching account.

claude-personal   # log in with personal account
claude-work       # log in with work account

Both accounts stay logged in across reboots until you /logout.

Migrating existing history into a profile

Each profile starts empty, so --resume and the up-arrow prompt history won't show anything you did before the split. If you want your existing ~/.claude/ history available in (say) the personal profile, copy it over once:

mkdir -p ~/.claude-personal
cp -R ~/.claude/projects     ~/.claude-personal/
cp    ~/.claude/history.jsonl ~/.claude-personal/
cp -R ~/.claude/file-history ~/.claude-personal/
cp -R ~/.claude/skills       ~/.claude-personal/
cp -R ~/.claude/agents       ~/.claude-personal/

What each one is:

projects/ — per-project conversation transcripts (--resume reads these)
history.jsonl — the input prompt history you cycle through with up-arrow
file-history/ — per-session file edit history
skills/ — your user-defined skills (slash commands, etc.)
agents/ — your user-defined subagents

Optional: copy settings.json if you've customized hooks, permissions, theme, or model — those carry over cleanly. Skip plugins/, cache/, paste-cache/, ide/, debug/, backups/, sessions/, session-env/, shell-snapshots/, telemetry/, and any auth/credential files — those are profile-specific and shouldn't be shared.

Safety net: prompt on bare `claude`

To avoid accidentally launching the wrong profile when you absent-mindedly type claude, shadow it with a fish function that forces a choice. Save as ~/.config/fish/functions/claude.fish:

function claude --description "Prompt for personal/work profile, then run claude"
    echo "Which Claude profile?"
    echo "  1) personal"
    echo "  2) work"
    read --nchars 1 --prompt-str "> " choice
    echo

    switch $choice
        case 1 p
            CLAUDE_CONFIG_DIR=$HOME/.claude-personal command claude $argv
        case 2 w
            CLAUDE_CONFIG_DIR=$HOME/.claude-work command claude $argv
        case '*'
            echo "claude: invalid choice '$choice' — expected 1/p (personal) or 2/w (work)." >&2
            return 1
    end
end

command claude bypasses the function so it doesn't recurse into itself. Anything other than 1/p/2/w prints an error to stderr and exits non-zero — no silent fall-through into a default profile. The aliases (claude-personal, claude-work) still work and skip the prompt — useful in scripts.

Notes

MCP servers are per-profile. If you configure an MCP server in one profile, you'll need to configure it again in the other.
Plugins, agents, hooks, and settings are per-profile for the same reason.
Existing ~/.claude directory is untouched and unused once the wrapper is in place — every claude invocation now goes through one of the two profiles.

Verifying

ls -la ~/.claude-personal ~/.claude-work