Enable and read the dashboard
This is an integrated optional module; it does not require running the standalone Mr. Usage app. Frog’s model-provider connections are separate from the coding-tool logins used here. An API key configured for a writing rule does not grant subscription-plan information.
- Turn on Mr. Usage in Features. It is off by default; enabling adds its separate menu-bar item and starts collection.
- Click the usage readout for its popup, or open Mr. Usage in the sidebar. Select Claude or OpenAI; this also selects the menu-bar provider.
- For Claude, choose a login source: Claude Code, Claude Desktop, OpenCode, Pi, or Automatic. Claude Code, OpenCode, and Pi use an existing login from that tool; Claude Desktop reads the limits the Claude app saves on this Mac and sends no request, but the Claude app saves them only now and then; Automatic shows the newer reading and is the dependable choice with a Claude Code login.
- Read available plan-limit percentages, resets, status, and credits. Use Refresh for an update or Open dashboard for the full view.
Several Claude accounts
When Frog finds a second Claude account on this Mac, such as another Claude app profile or Claude Code folder, the popup and the Mr. Usage page offer to show each one, with Show both and Not now. The same choice is the Show each Claude account switch on the Mr. Usage page. It is off by default.
With the switch on, the Accounts list names each account. Give it a label, include it or leave it out, or use Add folder… for a folder Frog did not find. Choose Each account to see the limits side by side or Combined for one set weighted by plan size. With Each account, the menu bar shows each account’s Session or Week limit, such as A 37% B 81%.
Frog reads the limits each Claude app profile saves on this Mac; an account known only from the Claude app causes no request. Live readings need a Claude Code sign-in per account, and Set up live readings… explains it before Terminal opens. When an account’s own Claude Code login has run out, its row offers Sign in to Claude Code…, which opens Terminal in that account’s folder. Labels and added folders stay on this Mac.
Token charts, model costs, and credits
Choose Last 24 hours, Last 7 days, or Last 30 days and a metric: API cost, Uncached input, Output, Cache write, or Cache read. The dashboard shows a chart, total, and per-model breakdown. The chart draws the period’s running total as a line, with each day’s or hour’s amount as a thin bar below it; move the pointer over it for exact values. The popup shares the selected metric and period.
Local Claude Code, Codex, OpenCode, and Pi records supply supported Claude and OpenAI token activity. The readers deduplicate supported message records rather than treating every transcript row as a new request. Missing logs or incomplete tool records can leave periods empty.
API cost is an estimated list-price equivalent, not your subscription charge or an invoice. Models with no known price are identified so missing pricing is visible. Claude extra usage and OpenAI additional-usage credits retain their own provider units rather than being conflated with the token chart’s cost estimate.
For OpenAI, choose This Mac or All devices as the activity source. All devices reconciles available complete account days with local records. Account totals can lag and token splits may be estimated; this is not a complete real-time log of every device.
Choose how the menu-bar item looks
Under Menu bar, at the bottom of the Mr. Usage page, each tile shows the item as it will look in your menu bar; click one to use it. Numbers is the readout with percentages and small bars. Steps lights five steps of 20% each, with a partly reached step drawn faint. Bars fills a smooth track, and Columns stands the meters side by side. Without percentages, the item takes less room and does not change with every point.
Colors sets Clean, which follows the menu bar like the system’s own icons; Traffic light, green that turns amber from 70% and red from 90%; or Spectrum, which runs from green to red along each meter. Each limit window, such as 5h and 7d, or each Claude account, such as A and B, gets its own row. Without a plan reading, the item still shows the period’s API estimate as text.
Collection continues with the dashboard closed
While enabled, Mr. Usage keeps collecting local activity and polling provider limits even when the dashboard and popup are closed, hidden, or minimized. The separate menu-bar item shows up to two available limit windows; when limits are unavailable, it shows the selected period’s API estimate.
Disable Mr. Usage in Features or quit Frog to remove the item, close its popup, and stop collection. Cancellation reaches requests, readers, and aggregation; late replies are discarded. Rate-limit cooldowns and backoff remain in force, including for Refresh.
Portable display preferences
Provider, period, metric, activity source, Claude login-source choice, the several-accounts switch, view, and menu-bar limit, and the menu-bar design and colors are shared between the popup and dashboard and saved in preferences.toolkit.usage. These display choices are included in configuration export/import. Credentials, account details, readings, and polling cooldowns are not exported; another Mac still needs its own originating-tool login.
Credentials stay with the originating tools
Claude usage requests use the selected tool’s existing login. OpenAI plan requests use available ChatGPT credentials from Codex or OpenCode. Frog reads those credentials for the provider request, does not export them, and never uses another tool’s refresh token. When Claude Code’s login has run out, Frog starts Claude Code in the background so that it renews its own login; that sends one minimal request on your Claude plan, at most every 30 minutes and less often while it fails.
If Claude Code can no longer renew its login, choose Sign in to Claude Code… in the popup or dashboard: Terminal opens Claude Code’s own sign-in, and Frog picks up the new login. For another tool, sign in again through that tool, then return to Frog. API-key-only logins or a credential store the integration does not support may not expose subscription limits; available local token records can still be useful.
Provider usage endpoints receive authenticated usage requests, not transcripts, prompts, notes, or dictation recordings. The integration sends no Frog telemetry and includes no leaderboard or sharing client. Last-good Claude limits, metadata, and request cooldowns use the separate local xyz.beffa.frog.usage defaults domain, outside config.json and the standalone app’s settings.
When limits or activity are missing
Check the selected provider and Claude login source, whether the originating tool is signed in, and whether it has recorded activity in the selected range. Read the popup or dashboard status for authentication or endpoint errors. An expired login is renewed by the tool that owns it; Frog asks Claude Code to do so and offers Sign in to Claude Code… when that is no longer possible.
A last-good or log-derived limit snapshot can remain useful when live requests fail, but it is not a fresh response. Available reset windows and credits depend on the provider and account. Keep the feature enabled and allow any reported cooldown to elapse.