English | 简体中文

See your Claude usage in the tray. Hit the limit, keep working on your next account. One click, no email code.
A tray tool for the Claude desktop app, for one account or several.
Several accounts: it swaps the signed-in account in about ten seconds, carries your Claude Code chats across
and shows the live usage of every account side by side.
One account: it is a usage monitor. The tray icon is a ring that fills with your 5-hour usage,
and the menu shows your 5-hour and weekly usage with the time each one resets.
Your sign-ins never leave your machine. Tokens stay encrypted exactly as Claude stored them,
and the only server they ever go to is api.anthropic.com.

Unofficial. Not affiliated with or endorsed by Anthropic.
The problem
Claude only shows your usage when you go looking for it. To see how much of your 5-hour or weekly limit is left, you open the usage page in Claude's settings, and the number is out of date the moment you close it.
With two accounts, because one runs out, it gets worse. Switching between them in the desktop app goes like this:
- Sign out. That throws the sign-in away. Next time you need it, you start from zero.
- Sign in with the other address and wait for the email code. With a private relay address, that means opening another app to find it.
- Find your work. Claude files Claude Code chats per account, so the chat you were in the middle of is gone from the sidebar. The conversation still exists on disk; the app just does not list it.
- Guess which account has room left, because usage only shows for the one you are signed in to.
Three of those four steps are pure overhead, and the fourth is a guess.
What it does instead
Keeps every sign-in. Each account's sign-in is saved when you switch away from it and put back when you switch to it. You type an email code once per account, ever, and not again until Anthropic ends that sign-in.
Switches in one click. Right-click the tray icon, pick an account. Claude quits, the sign-in is swapped, Claude starts again. Everything else in Claude's settings stays exactly as it was.
Brings your chats along. Before Claude starts, the switcher copies the chat list entries of your other accounts into the one you switch to. Open the chat you were in and keep going. Projects come along too, so chats stay grouped in them. Chats you deleted stay deleted. Scheduled tasks are shared as well, so each one runs once in whichever account is signed in, not again after every switch.
Shows your usage at a glance, with one account or many. The tray icon is a ring that fills with the active account's 5-hour usage: green, orange from 75%, red at your switch threshold. Hover it for the account and its numbers. The menu shows 5-hour and weekly usage for every saved account, live, with when each limit resets. A future 5-hour reset is shown even below the switch threshold, for example 5h 20%, resets 21:20. At or above the threshold it reads 5h 98% until 21:20. Times use a 24-hour clock in your local timezone.
Picks up where Claude stopped. A switch restarts Claude, and a chat that was working at that moment would just stand still. The switcher notices which chats were in the middle of a reply and asks in a small window which of them should go on. Claude then continues the checked ones on the new account by itself.
Can switch on its own. Off by default. Turn it on and the switcher moves Claude to the account with the most room left once the active one passes 96%, or 90, 94, 98 or any whole percentage you type in under Custom…. A second toggle moves it to an account whose weekly limit is about to reset mostly unused, so that usage is not lost.
Speaks Chinese. When Windows, macOS or Linux shows Simplified Chinese first, the menu and windows are in Chinese too. See Language.
Measured
The first real run, on Windows 11 with the Microsoft Store version of Claude, from the switcher's own log:
| Step | Time |
|---|
| Add a second account, including typing its email code | 44 s |
| Detect the new sign-in, import the missing chats, restart Claude | 9 s |
| Switch accounts, from the click until Claude starts again | 9 s |
Round trip of Claude's real config.json through the switcher | byte-identical |
Install
Download the file for your platform and run it. No installer, no runtime, one file.
Then turn on Start at login in its menu.
The builds are not code-signed, so the first start gets one warning:
Every download has a .sha256 file next to it, so you can check it is the file CI built. Signing plans and status: Code signing policy. What the switcher sends where: Privacy policy.
First run
- Start the switcher while you are signed in to Claude. It picks up that account within a minute.
- Choose Add account…. Claude restarts on its sign-in screen. Your first account is kept.
- Sign in with the second account. The switcher notices, restarts Claude once more, and the chats of your other account show up.
From then on, click an account to switch. With only one account, step 1 is all there is: the ring and the menu show its usage from then on, and you can add a second account any time.
> [!IMPORTANT]
> Never use Sign out in Claude to change accounts. Signing out can end that sign-in on Anthropic's side, and the switcher's saved copy stops working. The account then shows "signed out, add it again". Use Add account… to sign it in once more.
Auto-switch
Three toggles in the menu, all off until you turn them on:
- Auto-switch moves Claude to the account with the most room left once the active one is nearly full.
- Auto-switch to use up expiring weekly moves Claude to an account whose weekly limit resets soon with most of it unused, so that usage is not lost.
- Auto-switch on a model's weekly limit too counts the weekly limit of one model, such as Fable, as a limit of its own. The menu always shows that limit next to 5h and week when the plan has one. With the toggle on, Auto-switch also acts when it is nearly full, and both kinds of switch only go to an account with room in it. With several such limits on one account, the fullest counts.
A switch restarts Claude. A reply that is being written at that moment stops, and Claude picks it up again on the other account, see Chats that were working.
Anthropic's terms do not forbid having more than one account. They do forbid "bypassing any of our systems or protective measures". Whether automatic switching fits how you use your accounts is your call, which is why all toggles are off until you turn them on.
When Auto-switch acts
All of these have to hold before the switcher acts:
| Rule | Why |
|---|
| The active account is at or above the threshold, 5-hour or weekly | 96% by default, so the switch lands before a reply gets cut off at 100% |
| Another account is at least 10 points below the threshold | No bouncing onto an account that is nearly full too |
| The active account was checked in the last 5 minutes, the other one in the last 15 | No decisions on stale numbers |
| The last switch was more than 10 minutes ago | No ping-pong |
| Claude is running | It never starts Claude behind your back |
When Auto-switch to use up expiring weekly acts
Weekly usage you do not spend before the reset is gone. With this on, the switcher moves Claude to another account when all of these hold:
| Rule | Why |
|---|
| The other account's weekly limit resets within 24 hours | That is the usage about to be lost |
| It has used 50% of its week or less | Worth a restart only when a good part is left |
| It is at least 10 points below the threshold, 5-hour included | It can actually take work right now |
| The active account is not in the same spot itself, resetting within 24 hours at 50% or less | Its own unused week counts just as much |
| You did not switch by hand in the last 5 hours | Your own pick wins |
| No Claude Code chat was active in the last 15 minutes | This switch is never urgent, so it waits instead of cutting off a reply |
The freshness, cooldown and "Claude is running" rules above apply too. When several accounts qualify, the one that resets first wins. With both toggles on, being nearly full comes first: the account with the most room wins over the one that resets first.
Once it has moved to an account, it keeps using that week until the reset, even after more than half of it is used. If that account's 5-hour limit fills up and the normal auto-switch moves away, it comes back once the 5-hour window resets. Switching by hand ends this.
After any automatic switch, from either toggle, the top of the menu says why for 12 hours, for example "Auto-switched at 03:12 to use up this account's week before it resets 18:00".
Chats that were working
Claude has to quit for a switch, so every Claude Code chat that is writing a reply or running a tool stops. Before Claude starts again, a window lists those chats:
- Each chat shows its title and project folder, all checked.
- Continue lets Claude carry on with the checked ones. Don't continue, Escape or closing the window leaves all of them where they stopped.
- Nobody answering means all of them go on after 30 seconds, so an automatic switch at night does not leave work standing. Touching a checkbox stops the countdown.
When Claude is back, it sends each checked chat "The app was quit while you were working. Please continue from where you left off." and the chat goes on in the account you switched to. That is Claude's own way of continuing after a restart; the switcher only tells it which chats to continue.
How the switcher knows which chats were working
Right before Claude quits, the switcher reads the end of every Claude Code conversation written to in the last 30 minutes. A chat counts as working when its last message is a prompt or tool result Claude has not answered yet, or a tool call. A chat that finished its reply but still has a background agent at work counts too, because the restart stops that agent and nothing would start the chat's next turn: the agent's note in subagents/agent-*.meta.json says "requestShape": "background" and has no "finished": true yet and its transcript was written to in the last 30 minutes. Any other finished reply, or a reply you stopped yourself, does not count. A chat that only waits for a recurring job it set for itself with CronCreate does not count either: such a job leaves no trace on disk and ends when Claude quits, so after a switch the chat has to be started again by hand.
For each chat you keep checked, the switcher writes the mark Claude itself leaves when it quits during a reply (interruptedByQuitAt) into that chat's entry of the account you switch to, and removes the mark from the other accounts' copies, so the chat is not continued twice. Claude continues marked chats for 24 hours after the restart; a chat it cannot reach in that time simply stays where it stopped.
On Windows the window is drawn with WebView2, which comes with Windows 11 and current Windows 10. Without it, a plain Windows dialog asks the same question. On macOS it uses WebKit; like the rest of the macOS build, this has not been tested on a real Mac yet. On Linux it is a zenity or kdialog dialog.
Claude Code in the terminal
Off by default. Turn on Switch Claude Code in the terminal too and every switch, by hand or automatic, also moves the claude command to the same account, so the app and the terminal stay on one account. Windows and Linux only for now: on macOS Claude Code keeps its sign-in in the keychain, and the switcher does not touch it there.
Once per account, sign Claude Code in: switch the app to that account, then run /login in Claude Code (or claude auth login) and sign in with the same account. The switcher keeps a copy of that sign-in and puts it back whenever you switch to that account. Until an account has one, the top of the menu says so.
- Change accounts in Claude Code with
/login directly, not /logout first. /logout revokes the sign-in, which also ends the copy the switcher keeps.
- A running
claude reads the changed sign-in on its own, no restart needed.
- If Claude Code is signed in with an account the switcher does not have, it is left as it is.
ANTHROPIC_API_KEY, CLAUDE_CODE_OAUTH_TOKEN and the other credentials Claude Code ranks above /login win over the switched sign-in.
- The Code tab inside the desktop app always uses the app's own account and needs none of this.
How it works
The switcher saves and restores the sign-in of each account, copies the chat list, projects and scheduled tasks between accounts, and reads usage from api.anthropic.com. It backs up Claude's files before every switch and leaves the rest of Claude's settings alone.
Which files, and what happens to them
Claude keeps its sign-in in two places inside its data folder:
| Platform | Data folder |
|---|
| Windows, Microsoft Store app | %LOCALAPPDATA%\Claude-Data\claude_ since 2.42304, before that %LOCALAPPDATA%\Packages\Claude_\LocalCache\Roaming\Claude |
| Windows, direct download | %APPDATA%\Claude |
| macOS | ~/Library/Application Support/Claude |
| Linux, official Claude Desktop | ~/.config/Claude |
| What | Where | What the switcher does with it |
|---|
| Sign-in tokens | config.json: oauth:tokenCacheV2, oauth:tokenCache, lastKnownAccountUuid | Saves and restores these three entries, leaves every other entry untouched |
| Web session | Network/Cookies | Saves and restores the file while Claude is closed |
| Chat list | claude-code-sessions///local_.json | Copies missing entries in, refreshes older ones, never deletes |
| Deleted chats | deleted_ next to the entries | Reads them, so a deleted chat is never copied back |
| Projects | local-agent-mode-sessions///spaces.json | Adds missing projects, refreshes older ones, never deletes. Leaves out the claude.ai project link of the other organization, and a project you deleted in an account is not copied back |
| Scheduled tasks | claude-code-sessions///scheduled-tasks.json | Gives every account the same list. The account you switch away from has your latest edits, and the run times of all copies are merged, so a slot that ran in one account does not run again in the next. A task you delete in one account goes away in all of them. The same task set up in two accounts (same name and schedule) is kept once, the copy whose prompt you edited last |
| Chats stopped by a switch | interruptedByQuitAt in their local_.json | Sets Claude's own quit mark on the chats you chose to continue, removes it from the other accounts' copies and from chats you unchecked |
| Conversations | ~/.claude/projects | Reads the end of the ones written to in the last 30 minutes to see which chats are mid-reply. Never writes. They are shared by all accounts already |
| Claude Code sign-in, only with the terminal option | ~/.claude/.credentials.json: claudeAiOauth and the other entries of the claude.ai account. ~/.claude.json: oauthAccount. Both inside CLAUDE_CONFIG_DIR when that is set | Copies the current entries to the switcher's folder, puts in the saved ones of the new account and clears that file's account caches, the way /login does. Holds the locks Claude Code takes for its own token refresh meanwhile. Leaves MCP server tokens and every other entry alone. Between switches it only reads them, to keep the copy of the signed-in account current |
Quitting Claude. On Windows the switcher tells Claude's windows that Windows is closing the app, the way an installer does, and Claude quits within a second or two. A plain close request would only hide Claude in the tray. If Claude is still running after 20 seconds, it is ended.
Before each switch, config.json and the cookie store are backed up. The last 10 backups are kept. If any step fails, the backup goes back and the error shows at the top of the menu. With the terminal option on, both Claude Code files are backed up to cli-backups before they change, the last 10 kept, and a failed swap puts back what it changed.
Usage. The saved token cache is decrypted in memory with Claude's own key (DPAPI on Windows, the keychain on macOS, the desktop keyring on Linux). The switcher then asks api.anthropic.com for each account's profile and usage, the same endpoints Claude and Claude Code use themselves. The active account is checked every two minutes, every minute from 85%, the others every ten. That endpoint is rate-limited per account and Claude polls it too, so after a 429 the switcher waits as long as the server asks, or backs off up to 30 minutes, and keeps showing the last numbers meanwhile. Claude's tokens last about a month, so a saved account keeps showing live usage without being signed in.
Updates. Once a day the switcher asks GitHub for the version number of the latest release. Nothing about you or your accounts goes along. When a newer version is out, the top of the menu says so. A click downloads it from the GitHub release, checks it against the published SHA-256, puts it in place of the running file and restarts on it. A download the switcher makes itself gets no SmartScreen or Gatekeeper warning. If that fails, the download page opens instead. Starting a newer version by hand also works: it asks the running one to quit and takes over. With winget, winget upgrade SnlperStripes.ClaudeAccountSwitcher does the same. To never contact GitHub, quit the switcher, add "noUpdateCheck": true to state.json and start it again.
Started from inside Claude. Claude closes the programs started from inside it, a Claude Code terminal for example, when it quits, and older Store versions also show them a redirected AppData folder. The switcher notices when it was started that way and starts itself again through Explorer, so it sees the real files and survives Claude quitting.
Your data
| |
|---|
| Saved sign-ins, state, backups, log | ~/.claude-account-switcher (%USERPROFILE%\.claude-account-switcher on Windows), Open data folder in the menu. On Windows, webview in there is the WebView2 profile of the window that asks which chats continue |
| Tokens | The app's are stored exactly as Claude encrypted them, never written in plain text, never sent anywhere but api.anthropic.com |
| Claude Code tokens, only with the terminal option | Kept the way Claude Code keeps them itself on Windows and Linux: plain JSON readable only by your user, in accounts//cli.json and cli-backups. Never sent anywhere and never refreshed by the switcher. Turning the option off stops all of it; delete those files to remove the copies |
| Network | api.anthropic.com for profiles and usage. api.github.com once a day for the latest version number, off with "noUpdateCheck": true in state.json. github.com only when you click to update |
| Remove an account | Quit the switcher, delete its folder under accounts/ and its entry in state.json |
| Uninstall | Turn off Start at login, quit, delete the binary and the data folder |
Language
The menu and windows follow the first display language of the system: Simplified Chinese (zh-CN, zh-SG, zh-Hans) gets Chinese, everything else English. On Linux the language comes from the LANGUAGE and then LANG environment variables, so LANG=zh_CN.UTF-8 gives Chinese. The log stays in English.
To pick one for good, quit the switcher, add "language": "zh-CN" or "language": "en" to state.json in the data folder and start it again. It survives updates and the start at login.
Command line
claude-account-switcher -status
Prints every saved account with its usage and exits.
claude-account-switcher -status -json
Prints the same as one JSON object, with reset times in RFC 3339, for scripts and status bars. A limit without a reset time, such as a 5-hour window that has not been used yet, has no reset field. -json on its own does the same.
claude-account-switcher -version
Prints the version and exits.
Build
Go 1.26 or newer.
go test ./...
go build -ldflags "-H=windowsgui -s -w" -o claude-account-switcher.exe . # Windows
CGO_ENABLED=1 go build -o claude-account-switcher . # macOS
CGO_ENABLED=0 go build -ldflags "-s -w" -o claude-account-switcher . # Linux
Dependencies beyond the standard library: fyne.io/systray for the tray icon, and for the window that asks which chats continue jchv/go-webview2 on Windows and webview/webview_go on macOS. On Windows also golang.org/x/sys. On Linux godbus/dbus reads the sign-in key from the desktop keyring.
FAQ
Can the Claude desktop app be signed in to two accounts at once?
Not on its own. The app holds one sign-in at a time, and switching means signing out and waiting for an email code. The switcher keeps a saved sign-in for every account and swaps between them in about ten seconds.
How do I switch Claude accounts without signing out?
Install the switcher, add each account once with Add account…, then pick an account from the tray menu. You never use Sign out again.
What happens to my Claude Code chats when I switch accounts?
They come along. The switcher copies the chat list and projects of your other accounts into the one you switch to, so the chat you were in shows up in the sidebar and you can keep going.
What happens to a reply Claude is writing when I switch?
It stops, because Claude has to restart. Before Claude starts again, the switcher asks which of the stopped chats should go on, and Claude continues those by itself on the new account. See Chats that were working.
How do I see the usage of all my Claude accounts at once?
The tray menu shows 5-hour and weekly usage of every saved account, live, with when each limit resets. claude-account-switcher -status prints the same in a terminal.
Is it useful with only one Claude account?
Yes, as a usage monitor. Start it while you are signed in to the Claude desktop app and it picks the account up by itself, no setup. The tray ring and the menu then show that account's 5-hour and weekly usage. The auto-switch toggles only act once a second account is added.
How do I see my Claude usage in the Windows system tray?
Start the switcher. Its tray icon is a ring that fills with your 5-hour usage and turns orange from 75%, red at the threshold under Auto-switch at (96% by default). Hover it for the numbers, open the menu for the 5-hour and weekly limits with reset times. Windows may put a new tray icon into the overflow area, so pin it to the taskbar in the taskbar settings to keep it in sight.
Can I see my Claude usage in the macOS menu bar?
The macOS build puts the same ring in the menu bar, but nobody has tested it on a real Mac yet. A test report helps.
How do I track the Claude 5-hour limit and weekly limit?
The menu line of each account shows both, for example 5h 42%, resets 19:00 · week 18%, resets Mon 09:00. The signed-in account is checked every two minutes, every minute from 85%. These are the same numbers Claude and Claude Code read from api.anthropic.com for your plan.
Do I need the Claude desktop app?
Yes. The switcher reads the sign-in of the Claude desktop app, on Windows or macOS. If you only use the claude command in a terminal, it has nothing to read.
What can I do when I hit the Claude usage limit?
Switch to an account that has room left, or let Auto-switch do it for you shortly before the limit.
Does it work with the Microsoft Store version of Claude?
Yes. It finds the Store app's real data folder and works with the direct download too. Windows 10 and 11.
Does it work on macOS?
It builds for macOS (universal) in CI, but nobody has tested it on a real Mac yet. A test report helps.
Does it work on Linux?
Yes, as a beta since v0.4.0. The port's author ran it on Ubuntu 24.04 (x64); other distributions and desktops have not been tried yet. Builds for x64 and arm64 come with every release. It needs the official Claude Desktop for Linux beta, see Install. A test report helps.
Does it switch the Claude Code CLI in the terminal?
Yes, if you turn it on, on Windows and Linux. With Switch Claude Code in the terminal too checked, every switch also moves the claude command to the same account, so the app and the terminal stay on one account. It is off by default and each account needs one /login in the terminal first, see Claude Code in the terminal. Not on macOS yet.
Do my sign-ins leave my machine?
No. Tokens stay exactly as Claude and Claude Code stored them, and the only server they ever go to is api.anthropic.com. The only other request is a daily check on GitHub for a new version, which sends nothing about you and can be turned off, see How it works.
How do I update?
The menu says so when a new version is out; click it and the switcher updates itself and restarts. Or download the new file and start it, it replaces the running one. With winget, run winget upgrade SnlperStripes.ClaudeAccountSwitcher. Your accounts and settings stay in the data folder.
Troubleshooting
Where is the log?
The switcher writes switcher.log in its data folder, ~/.claude-account-switcher (%USERPROFILE%\.claude-account-switcher on Windows). Open data folder in the tray menu opens that folder. When the log is larger than 1 MB at start, it starts over.
How do I see every account at a glance?
Run claude-account-switcher -status. It prints one line per saved account with its 5-hour and weekly usage, marks the signed-in account with *, and exits. It only reports; a switch that is due is left to the tray.
A switch failed after a Claude update.
Before a switch, the switcher copies config.json and the cookie store into the backups/ folder in the data folder. If the change to Claude's files fails, it restores that backup and shows the error at the top of the tray menu. The same error is in switcher.log. The last 10 backups are kept.
What makes a useful bug report?
The Claude desktop version (Help → About), the switcher version (claude-account-switcher -version prints it, and the first line of switcher.log names it), the last 30 lines of switcher.log around the problem, and what you did, what you expected and what happened instead. Do not paste tokens, cookies or config.json.
The account did not change after switching.
A switch closes Claude, replaces the sign-in while it is closed, and starts Claude again. A window that stays open never changes account. If the menu item of the account is already checked, clicking it does nothing. -status marks the account the switcher reads from Claude's config.
The model did not change after switching.
The switcher does not change the model. It saves and restores only the three sign-in entries in config.json and the cookie files; every other entry in config.json is left as Claude had it.
Caveats
This relies on how the Claude desktop app stores its files today, not on a published interface. An app update can change that. When it does, the switcher fails loudly, restores its backup and says so in the menu. It does not guess.
Contributing
Bug reports, ideas and pull requests are welcome, see CONTRIBUTING.md.
Got a Mac? A macOS test report helps a lot, even if everything just worked.
Running Linux? A Linux test report helps the same way.
Found a security problem? Please report it privately, see SECURITY.md.
If the switcher saves you time, a star helps other Claude users find it.
License
Apache 2.0. Use it, change it, ship it, sell it.
Keep the credit. If you pass this on, changed or not, include the NOTICE file, which names the author and links back here, and say what you changed.