c
One command to pick a Claude Code account and launch it, showing each account's 5-hour and weekly usage with the time until reset. A single dependency-free file, published on npm as @tonoid/c.
One command to pick a Claude Code account and launch it, showing each account's 5-hour and weekly usage with the time until reset. A single dependency-free file, published on npm as @tonoid/c.

Anyone running Claude Code across several accounts ends up with a drawer of shell aliases: one CLAUDE_CONFIG_DIR=... claude per account, doubled for the ones that skip permission prompts, doubled again for the ones that open a worktree. The aliases say nothing about which account still has quota. You pick one, work for ten minutes, hit the 5-hour limit, quit, and try the next. Checking first means launching a session and running /usage inside it, which is exactly the thing you were trying to avoid.
Type c and you get a table: every logged-in account, its plan, its 5-hour and weekly usage as a percentage, and the time until the window resets. Arrow keys move the selection, enter launches, and the menu redraws in place rather than scrolling a new copy past you. Rows sort most-recently-used first, so enter is always the account you were just in. There is nothing to register: any ~/.claude* directory holding credentials is an account, so a manual CLAUDE_CONFIG_DIR login shows up on its own. Defaults for --dangerously-skip-permissions and --worktree are remembered and toggled from the menu, and skipped when you type the flag yourself.
Developers running Claude Code across more than one account: a personal plan and a work plan, or a team seat alongside their own. Anyone who has written those aliases by hand and wants the quota picture before committing to a session rather than after.
CLAUDE_CONFIG_DIR set for the child process only, so two terminals can run two accounts at oncenode:test suite against throwaway HOME directories, so it never reads real accounts and never calls the networkLive and MIT-licensed, published as @tonoid/c. Code generated with Claude Code, then tested and reviewed manually. The usage endpoint is undocumented: if it changes, the usage columns show an error and picking and launching still work. Linux only for now, since macOS keeps Claude Code credentials in the Keychain rather than on disk. Open to issues and PRs.