Anthropic Re-allows OpenClaw-style Claude CLI Usage

Anthropic announces that it has re-allowed Claude CLI reuse in OpenClaw and the use of claude -p. Until a separate policy change is issued, OpenClaw will treat this usage method as an officially supported integration.

For production environments, an Anthropic API key is still presented as the clearest choice. If you already have a Claude CLI login on the host, you can reuse that account as-is.

Additionally, OpenClaw supports the following features for Anthropic integration.

• Default adaptive thinking for the Claude 4.6 series
• /fast mode and service_tier mapping for the Anthropic API
• prompt caching support, with short (5 minutes) applied by default when using an API key
• 1M context window beta option for supported models
• Bundled claude-cli backend support

However, some features are limited depending on the auth method.

• prompt caching is API-only, and cache settings do not apply under legacy token auth.
• 1M context has a beta gate, and under legacy token auth (sk-ant-oat-*), context-1m-* requests may be rejected.
• service_tier is injected by OpenClaw only when sending directly to api.anthropic.com.

In summary, OpenClaw has organized and provided Claude CLI reuse, API key, caching, and large context in line with Anthropic's latest CLI usage scope.