Skip to main content

Anthropic

Header-only rate-limit probe for the Anthropic API. OpenUsage sends a minimal request to /v1/messages and reads RPM/TPM headers from the response.

At a glance​

  • Provider ID — anthropic
  • Detection — ANTHROPIC_API_KEY environment variable
  • Auth — API key
  • Type — API platform (header-only rate limits)
  • Tracks:
    • RPM and TPM rate limits (limit, remaining, reset)
    • Auth status

Setup​

Auto-detection​

Set ANTHROPIC_API_KEY. OpenUsage registers the provider on next start.

Manual configuration​

{
  "accounts": [
    {
      "id": "anthropic",
      "provider": "anthropic",
      "api_key_env": "ANTHROPIC_API_KEY",
      "base_url": "https://api.anthropic.com/v1"
    }
  ]
}

Set base_url for proxies or compatible gateways.

Data sources & how each metric is computed​

OpenUsage sends one POST https://api.anthropic.com/v1/messages per poll cycle (default every 30 seconds in daemon mode). The body is minimal so Anthropic responds with HTTP 400, but the response headers carry rate-limit data and that is all this provider reads. The body is discarded.

Request headers:

  • x-api-key: $ANTHROPIC_API_KEY
  • anthropic-version: 2023-06-01
  • Content-Type: application/json

rpm — requests per minute​

  • Source: response headers
    • anthropic-ratelimit-requests-limit
    • anthropic-ratelimit-requests-remaining
    • anthropic-ratelimit-requests-reset
  • Transform: copied verbatim into the metric's Limit and Remaining. The reset string is parsed as RFC3339 and stored as Resets["rpm"].
  • Window: 1 minute.

tpm — tokens per minute​

  • Source: response headers
    • anthropic-ratelimit-tokens-limit
    • anthropic-ratelimit-tokens-remaining
    • anthropic-ratelimit-tokens-reset
  • Transform: same as rpm but for tokens.

Auth status​

  • Source: HTTP status code of the probe.
  • Transform: 401/403 → auth; 429 → limited; otherwise ok. The 400 that the empty-body probe triggers still carries valid rate-limit headers, so the tile reads ok.

What's NOT tracked​

  • Spend / cost. The API does not expose dollar figures or usage totals to API tokens, and there is no billing endpoint a key can authenticate against. Install Claude Code for token-level cost estimates from local session logs.
  • Per-model breakdown. The probe is a single request; the headers reflect your active tier, not a model-by-model split.

How fresh is the data?​

  • Polled every 30 s by default (data.poll_interval). Each poll is one request, no cache.

API endpoints used​

  • POST /v1/messages — header-only probe with anthropic-version: 2023-06-01

Caveats​

note

Anthropic's API does not expose spend or token-usage data to API keys. For full visibility install Claude Code, which reads local sessions and computes per-model costs.

  • Rate limits come from response headers and reflect the active tier.
  • The probe is a single minimal request per poll — negligible cost.

Troubleshooting​

  • Auth failed — verify ANTHROPIC_API_KEY and rotate if necessary.
  • Stale reset times — Anthropic rolls reset windows; the next poll picks up the new value.

Why is there no $ spend?​

The Anthropic API does not return spend or token-usage data on response headers, and there is no per-key billing endpoint we can authenticate against. The Claude Code provider closes that gap by reading on-disk session logs and multiplying token counts by published pricing.

  • Claude Code — local sessions, billing blocks, burn rate for the same models