Claude Code spend limit: how to cap what Claude Code can cost

The short version

• Console (API key billing): cap the auto-created Claude Code workspace on the Console's Workspaces page → select it → Spend limits. It's the only workspace that supports per-user monthly spend limits.
• Pro, Max, Ultra: the plan has usage windows; anything beyond them is paid from usage credits, which have a monthly spend limit in Settings → Usage on claude.ai.
• Team, Enterprise: the seat allowance is the default ceiling; with usage credits on, admins set spend limits for the organization, groups or members in Organization settings → Usage.
• Bedrock, Google Cloud, Microsoft Foundry: your cloud's budget controls, or per-developer caps through a Claude apps gateway.
• Per run: claude -p --max-budget-usd 5.00 "…" stops a print-mode run on an estimate; spend can pass it.

1. Find out which billing path you're on

Anthropic's Manage costs page (read 9 Nov 2026) is explicit that each developer is metered by the sign-in method they used, and an organization can mix them. Run /status in Claude Code to see the active credential. Watch for one trap: the error reference warns that a stray ANTHROPIC_API_KEY in your environment can route requests through an API key instead of your subscription, so a Max user can end up paying per token without noticing.

2. Console organizations: cap the Claude Code workspace

Per Anthropic's Workspaces docs, the first time a member signs in to Claude Code with a Console account, Anthropic creates a workspace called Claude Code and adds them to it. Claude Code mints a per-user API key there at sign-in; you can't create keys in it by hand, and a member's key stops working when they're removed.

1. Open platform.claude.com/settings/workspaces and select Claude Code.
2. In the panel, click Spend limits, set a monthly cap and the alert thresholds you want.
3. Optional: under Rate limits, pick the Claude Code workspace and cap requests or tokens per minute, so a busy afternoon of agents can't starve production traffic.

Things to know before you rely on it:

• A workspace limit can be lower than the organization's limit, never higher, and organization-wide limits always apply.
• Workspace spend limits are monthly caps; the docs don't describe a daily one, so a bad day can still spend a large part of the month's cap.
• Don't archive the Claude Code workspace to “turn it off” unless you mean it: archiving disables Claude Code sign-in through Console billing for the whole organization and can't be undone.
• Your scripts and apps should not live in the Default Workspace either, because the Default Workspace can't have a spend limit at all. See Anthropic Default Workspace limits.

3. Per-run caps: --max-budget-usd and --max-turns

For scripted or CI runs, the CLI reference lists two print-mode-only guards:

# stop this run once estimated spend reaches $5
claude -p --max-budget-usd 5.00 "fix the failing tests in api/"

# and/or cap the number of agentic turns
claude -p --max-turns 20 "fix the failing tests in api/"

What the docs say about --max-budget-usd:

• It checks Claude Code's client-side cost estimate, which can differ from your bill.
• Spend can pass the cap, so leave headroom. Subagent spend counts toward it.
• Totals restored from earlier runs with --continue or --resume don't count toward it.
• Once the cap is reached, spawning another subagent fails with Budget limit reached and background subagents are stopped (v2.1.217 or later).
• It doesn't apply to interactive sessions.

So use it as a seat belt for one run. The real ceiling is still the workspace, plan or cloud limit above.

4. Which message means which limit

Sources: Claude Code error reference and Anthropic rate limits. Note that a spend limit on any other workspace returns HTTP 400, not 429. If you run unattended jobs with CLAUDE_CODE_RETRY_WATCHDOG=1, recent versions stop at once on a spend-limit 429 instead of retrying it forever.

5. Watch the spend before a cap trips

• /usage shows the session's token use and a dollar figure computed locally at list price: an estimate, not a bill.
• Console: the Claude Code dashboard shows spend and accepted lines per member; the Claude Code Analytics API returns daily per-user metrics with an Admin API key.
• Any setup: OpenTelemetry export streams per-user token and cost metrics into your own monitoring.
• Usual causes of a surprise: long sessions never cleared, Opus left as the default, scheduled loops firing while idle, and agent teams (Anthropic says about 7× the tokens of a standard session when teammates run in plan mode).

Checklist

1. Run /status on every machine and confirm the credential is the one you expect (no stray ANTHROPIC_API_KEY).
2. Console: set a monthly spend limit and alert thresholds on the Claude Code workspace, and per-user limits if your team needs them.
3. Subscriptions: set the usage-credit monthly spend limit (or keep usage credits off).
4. CI and scripts: add --max-budget-usd with headroom and a --max-turns ceiling.
5. Keep your own app keys out of the Default Workspace, in a workspace with its own spend limit.
6. Have something outside Claude Code tell you the same day when organization spend is far above normal.

Want to check whether your Anthropic (and OpenAI / OpenRouter) limits would actually stop a runaway loop or a leaked key? The free AI key limit audit walks through them in a few minutes (no sign-up, no key needed). Related: runaway AI agent cost, workspace spend limits via the API.

Where KeyBark fits (and where it doesn't)

KeyBark reads your Anthropic organization's cost with an Admin key about every 15 minutes during the beta and emails you when a daily or monthly dollar threshold you set is crossed, or when today's spend jumps above your 7-day average. Claude Code usage billed to your Console organization is part of that organization cost, so a busy day of agents shows up there.

Limits, honestly: KeyBark watches the whole organization, not the Claude Code workspace or each developer, and it does not set Anthropic spend limits for you. It does not see Pro, Max, Team or Enterprise plan usage or usage credits, and it does not see Claude Code on Bedrock, Google Cloud or Foundry. Provider cost data can lag, and checks run on a schedule, so an alert can arrive after money has been spent. KeyBark is not a proxy and can't block a request. Set the limits above first; KeyBark is a second line of warning on top.

FAQ

Sources (read 9 Nov 2026): Claude Code Manage costs, CLI reference, Error reference, Claude apps gateway spend limits; Anthropic Workspaces and Rate limits. KeyBark is not affiliated with Anthropic.