Claude Code rolling 5h + weekly quota: it is four clocks, not two.

Claude Code charges four independent rolling buckets on every turn. five_hour is a 5-hour sliding window across all your Anthropic activity. seven_day is the overall 168-hour rolling cap. seven_day_opus is the Opus-only 168-hour cap. seven_day_oauth_apps is the 168-hour cap that fills only from OAuth clients, which means Claude Code and MCP traffic; claude.ai browser chat does not touch it. Each is a utilization float between 0.0 and 1.0 with its own resets_at timestamp. Any one at 1.0 returns the next 429. The endpoint is GET claude.ai/api/organizations/{org_uuid}/usage and the Rust struct that declares every field is at src/models.rs lines 18-28 in github.com/m13v/claude-meter.

Because /usage and the Settings page hero bar both surface seven_day (the overall weekly bucket), but the 429 follows whichever of the four buckets is highest. For a Claude Code-heavy account that does no browser chat, the bucket climbing fastest is usually seven_day_oauth_apps. It can sit at 0.94 while seven_day reads 0.62 because seven_day is a blended average across Claude Code, browser chat, and any other OAuth client traffic on your org. The Settings page does not render seven_day_oauth_apps as its own bar, so the bucket biting you is invisible until you open DevTools or run a tool that reads the full JSON.

Yes. The 5-hour clock rolls 5 hours after your first message in the rolling window. Each weekly clock rolls 168 hours after the first message of that weekly cycle, separately per bucket. A 5-hour reset does not advance the weekly clock, and a weekly reset does not unlock the 5-hour bucket. Both timestamps come back on the same JSON response as resets_at. claude-meter renders each as a relative duration so you can see which one bites first.

No. Anthropic doubled the rolling 5-hour limit on Pro, Max, Team, and seat-based Enterprise plans on May 6 2026. The weekly caps were not doubled. So Claude Code users on Max see roughly 2x the 5-hour throughput but the same weekly headroom. The most common surprise after May 6 is the 5-hour bar looking healthy while seven_day_opus or seven_day_oauth_apps quietly climbs past 90 percent earlier in the week than it used to.

Different ledgers. ccusage walks ~/.claude/projects/<project>/<session>.jsonl on your disk and totals input_tokens plus output_tokens against a price card. That gives you a faithful local token count and dollar estimate. The four utilization floats live on Anthropic's server, are weighted by peak-hour multipliers, factor in attachments and tool-call overhead, and include any browser-chat traffic the JSONL files never see. ccusage and a server-truth meter answer different questions on purpose. Run both.

Open claude.ai/settings/usage in Chrome. Open DevTools (Cmd+Option+I), switch to the Network tab, filter on /usage. Refresh the page. The fetch on page load returns the full JSON with every Window field. Click the Response tab to see the entire enforcement state. Whichever utilization is closest to 1.0 is the bucket that will 429 your next prompt. The fields you care about for Claude Code are five_hour, seven_day, seven_day_opus, and seven_day_oauth_apps.

It is your own session calling the same endpoint claude.ai/settings/usage already calls every time you open the page. claude-meter polls once per minute per browser, well below normal browsing traffic. The endpoint is undocumented, which means Anthropic can rename fields, not that they prohibit reading them. The Rust types in src/models.rs declare every Window field as Option, so when the JSON shape changes the next release patches the new fields in one line.

They convert, they do not vanish. With metered billing on, prompts past plan caps go through at standard API prices until you hit your monthly_credit_limit. When used_credits crosses that cap, GET /api/organizations/{org_uuid}/overage_spend_limit returns out_of_credits: true and a disabled_until timestamp. So you stop being rate-limited by a plan bucket and start being rate-limited by a dollar cap. Both are 429s. claude-meter renders the dollar cap as a separate Extra usage row with a BLOCKED suffix when it fires.

brew install --cask m13v/tap/claude-meter to install the macOS menu bar app, install the browser extension from github.com/m13v/claude-meter/releases, then visit claude.ai once. The extension hands the live claude.ai session to the menu bar app, so there is no cookie paste. Numbers refresh every 60 seconds and match claude.ai/settings/usage exactly because the source endpoint is the same.