Use Anthropic with OpenClaw

OpenClaw supports two Anthropic authentication paths: direct API access with usage-based billing, or reuse of a Claude CLI login on the same host.

Option 1: Anthropic API key

Run the interactive onboarding flow and choose Anthropic API key:

openclaw onboard

This is the clearest option for a shared or long-running Gateway because the credential and billing boundary are explicit. Store the key through the supported credential flow or a protected service environment, not in a public configuration file.

Option 2: reuse Claude CLI login

First confirm that Claude CLI is installed and signed in on the same host that runs OpenClaw:

claude --version
openclaw onboard

Choose Claude CLI during onboarding. This path depends on the local Claude CLI login and runtime compatibility. A container or remote Gateway does not automatically inherit credentials from the host user’s ~/.claude directory.

Discover and select a model

openclaw models list --provider anthropic
openclaw models status
openclaw models set <anthropic/model-id>

Use the exact anthropic/* reference reported by the current CLI. Rolling aliases and provider catalogs can change, so pin an exact model reference when reproducibility matters.

Verify the connection

Test a normal answer and a harmless tool call. If a newly created agent reports no Anthropic key, remember that credentials can be scoped per agent; configure that agent rather than copying secrets into its prompt.

Expected result

The provider appears in openclaw models status, the selected model is available, and a test turn completes without a missing-credential, expired-token, or provider-policy error.