cchost

Host multiple isolated Claude Code accounts on one machine via a single cc command. cc is a plain Node CLI on your PATH, so it runs the same from PowerShell, cmd, bash, zsh, or fish — there are no shell scripts or rc-file edits involved. Tested on Windows and Linux; macOS is expected to work but is not yet verified.

Each profile is a complete, isolated CLAUDE_CONFIG_DIR — credentials, session history, project state, plugins, and settings. Create one with cc add <name>; launch under it with cc use <name>.

Install

npm install -g cchost

Requires Node.js 18+ and Claude Code 2.1.140+ already installed. Get Claude Code from https://code.claude.com (npm install -g @anthropic-ai/claude-code).

Commands

Examples

cc add work                            # create the profile, do not launch
cc use work                            # first launch: Claude prompts for login
cc use work --resume                   # resume last conversation under "work"
cc use personal -p "summarize"         # one-shot prompt under "personal"
cc list
cc remove old-profile

cc list output:

Profile   LoggedIn  Email                 Storage
personal  true      personal@example.com  isolated
scratch   false     -                     isolated
work      true      work@example.com      shared

Migrating existing projects

If you've been using Claude Code without profiles, your history lives in ~/.claude/projects/. Pull it into a profile or the shared pool:

cc migrate default work        # copy ~/.claude/projects → ~/.claude-profiles/work/projects
cc migrate default shared      # copy ~/.claude/projects → ~/.claude-shared/projects
cc migrate shared personal     # copy ~/.claude-shared/projects → ~/.claude-profiles/personal/projects

This is a copy — the source is left untouched. If an entry already exists at the destination, the source entry is skipped (no overwrite). Credential files (.credentials.json, .claude.json) are never copied: each profile keeps its own auth state.

Shared history across profiles

If you want each profile to keep separate credentials but share project history (e.g. rotate accounts on the same project when usage limits hit):

cc link work
cc link personal

work/projects and personal/projects become links to ~/.claude-shared/projects. On Windows this uses directory junctions (no admin/Developer Mode required); on macOS/Linux, regular symlinks.

Only projects/ is shared. That's the directory Claude Code uses for resumable conversation transcripts (projects/<project>/<session>.jsonl). The sibling sessions/ folder holds transient per-process registry state — PID, working directory, status, timestamps — and is never linked, migrated, or copied across profiles. Sharing it would corrupt Claude's running-process bookkeeping. sessions/ stays per-profile under ~/.claude-profiles/<name>/sessions/.

cc link a b c links several profiles at once; per-profile failures are reported but don't abort the rest of the batch.

If two profiles share history and you launch them against the same project at the same time, both will write to the same session files (last write wins). Don't do that.

cc remove is safe with shared profiles — it deletes only the link, never recursing into the shared target.

Aliases (cc add flags) — EXPERIMENTAL

cc add work --email=work@example.com --org=Acme --name="Work Account" stores those values in ~/.claude-profiles/work/.cc-alias.json. On every cc use, they're written into the profile's .claude.json oauthAccount so cc list and Claude's own UI display them instead of the real underlying account.

EXPERIMENTAL. Claude Code may overwrite oauthAccount on server re-sync; we re-apply on the next cc use, but masking across all of Claude's UI surfaces has not been verified. Use cc list --original to see the real values from .claude.json.

How it works

cc is a Node CLI with zero runtime dependencies. For cc use it:

1. Resolves the profile directory under ~/.claude-profiles/<name>/
2. Applies any stored alias to .claude.json oauthAccount (best-effort)
3. Sets CLAUDE_CONFIG_DIR in its own spawn environment
4. Spawns claude with stdio: 'inherit' so the TUI works correctly
5. Forwards the child's exit code

The env var is set on the child process only — that's all claude needs.

Sensitive data

~/.claude-profiles/<profile>/ holds real OAuth credentials. Treat it like SSH keys: don't commit it, don't share it, don't sync it to public cloud storage. The .gitignore shipped with this project excludes profile directories from anywhere they might land. cc migrate and cc link never copy .credentials.json or .claude.json across profiles.

Troubleshooting

Claude Code does not appear to be installed Install it (npm install -g @anthropic-ai/claude-code) and make sure claude --version runs in a fresh terminal.

warning: Claude Code <version> is older than 2.1.140 Earlier versions had bugs where Windows read .claude.json from %USERPROFILE% regardless of CLAUDE_CONFIG_DIR, breaking isolation. Run claude update.

Profile "<name>" does not exist. Create it with: cc add <name> cc use does not auto-create profiles — that prevents typos like cc use wokr from silently creating a new, blank one. Run cc add <name> first.

Profile name rejected Names must be free of path separators, whitespace, .., leading dots, and reserved DOS names (CON, PRN, COM1…). Stick to letters, digits, dashes, underscores.

Old shell-script version was leaking keystrokes / leaking use work into the prompt That was a known bug in the cmd-based predecessor (%* ignored shift). This package forwards only the argv after <profile> to claude; the action and profile name themselves are consumed by cc.

License

MIT