How to cap Claude Code spending for a team

On a Team or Enterprise plan, each seat's allowance is the ceiling until usage credits are turned on; then you limit Claude Code spending for a team at Organization settings › Usage, per organisation and per member. On the API, set workspace spend limits in the Claude Console. On a cloud provider, use its budgets or a gateway. For scripts, add --max-budget-usd to each run.

There's no single switch, because Claude Code is billed four different ways. The controls for each come from Anthropic's docs and help pages as we read them on 22 October 2026 and again on 25 October 2026.

Which levers you have, by setup#

For scale, Anthropic's cost guide gives these figures: "across enterprise deployments, the average cost is around $13 per developer per active day and $150-250 per developer per month, with costs remaining below $30 per active day for 90% of users." Before a wider rollout it suggests a small pilot group, to set your own baseline (what Claude Code costs per developer breaks the numbers down).

On a Team or Enterprise plan#

A member's Claude Code use draws from their seat's allowance, the same one Claude chat and Cowork draw from. It resets on a rolling 5-hour window and on a weekly window, and its size depends on the seat tier, Standard or Premium (how big each allowance is). The spend report in org analytics appears once usage credits are on; use inside the allowance isn't metered in dollars.

With usage credits off, the seat is the cap and nobody spends more than their seat. A member who runs out waits for the window to reset, or types /usage-credits in Claude Code, which sends the organisation's admins a request. Spending starts only when you turn usage credits on, so members can keep going past their allowance. To cap them (Claude Help Center):

1. As an Owner or Primary Owner, open Organization settings › Usage.
2. Set a monthly spend limit for the whole organisation.
3. Set individual monthly limits for the members who need a different ceiling.
4. On a seat-based Enterprise plan, you can also set a limit per seat tier.

A member who reaches their limit has their usage credits paused until the end of the month. On Enterprise with usage credits on, the Spend Limits API does the same from a script: it sets a limit on each member and approves or denies their requests for a higher one, with a scoped Admin API key. Console organisations can't use it.

Fast mode runs Opus faster at a higher price per token, and on subscription plans it's paid from usage credits only, even when the member has plan usage left. Until usage credits are on, /fast answers "Fast mode requires usage credits". It's off for Team and Enterprise organisations until an Owner enables it at Organization settings › Claude Code. If you do enable it, "fastModePerSessionOptIn": true in managed settings makes every session start with it off, and people turn it on with /fast each time.

On the Claude Console (API keys)#

API organisations pay per token, so a cap here is a hard money limit. From the workspaces docs:

• On the Console's Workspaces page (platform.claude.com/settings/workspaces), select a workspace and click Spend limits to cap its monthly spending and set alerts at thresholds. A workspace limit can be lower than your organisation's, never higher, and you can't set one on the Default Workspace.
• A workspace called "Claude Code" appears the first time a member signs in to Claude Code with a Console account. Claude Code mints a per-user API key in it at sign-in, and nobody can create keys there by hand. It keeps Claude Code's usage apart from your other API work, and it's the only workspace that supports per-user monthly spend limits.
• Under Rate limits, pick the Claude Code workspace from the Workspace dropdown and click Edit next to a model to cap its requests per minute, input tokens or output tokens, so a busy afternoon of agents can't starve your production traffic.

For per-person numbers, the Console's Claude Code dashboard (platform.claude.com/claude-code) shows spend and accepted lines per member.

On Bedrock, Google Cloud or Foundry#

Here Claude Code bills to your cloud account. The cap lives in your cloud's budget tools, and Anthropic's analytics don't see this usage. Anthropic's docs list three ways to get a figure per developer:

• A Claude apps gateway you host. Its spend limits cap each developer per day, week or month, with defaults per group or for the whole organisation: each cap is a POST /v1/organizations/spend_limits with a scope, an amount in cents and a period, so "amount": "50000", "period": "monthly" is $500 a month. Daily caps reset at 00:00 UTC, weekly ones on Monday, monthly ones on the first. A developer over the cap gets a 429 reading, for example, spend limit reached (daily; resets 2026-08-08 00:00 UTC), and Claude Code warns them at 75% and 95% of it.
• An LLM gateway that tracks spend per key. The docs name LiteLLM and note that it's unaffiliated with Anthropic and not audited by them.
• OpenTelemetry export from each machine, which reports spend but enforces nothing.

Cap a single run with --max-budget-usd#

Scripts and CI jobs get a per-run cap from the CLI reference:

claude -p --max-budget-usd 5.00 --max-turns 30 "fix the failing tests in packages/api"

That run ends when Claude Code's estimate of its API spend reaches $5.00, or when --max-turns runs out at 30 agentic turns and exits with an error (there's no turn limit by default). --max-budget-usd works in print mode (-p) only. Subagents count toward it; at the cap, spawning another one fails with Budget limit reached and background subagents are stopped (Claude Code v2.1.217 or later). The check runs against Claude Code's own estimate, which can differ from your bill, and spend can pass the cap, so leave headroom. A --continue or --resume run starts its count fresh: totals restored from earlier runs don't count toward the cap.

Model and usage controls for admins#

Managed settings take an availableModels allowlist, which limits the models people can pick. On Team and Enterprise, each person's /usage screen shows what's driving their own use.

How Shuriken helps with spending#

Shuriken never bills model usage. Agents run on the plans and keys you add as sources, so every limit above still applies. On top of them:

• The Analytics page shows spend by source and by project, across every account and key you added (spend by account and project).
• Switch to API keys is off by default. Turn it on in Switch settings at the top of Sources and API keys take chats only once every subscription of that agent is out. Spend cap per fallback, on by default at $25, then stops the run. It counts the whole run for that agent's provider across every key it uses, not each key separately, and a subscription with room again ends the run (API keys as a last resort, with a spend cap).
• In Shared projects (beta), sharing a project (Share… in its ⋯ menu) gives that member a daily spend cap in dollars, $10's worth to start. They see what they've spent today against it, and their chats pause for the day once they reach it (why we built shared projects).