Anthropic has quietly confirmed what many in the AI agent community have been waiting to hear: OpenClaw-style Claude CLI usage is officially allowed. In a statement that should put months of uncertainty to rest, Anthropic staff told the OpenClaw team that their Claude CLI reuse and "claude -p" usage is sanctioned — at least for now. The bundled Anthropic claude-cli backend is fully supported in OpenClaw, and the project is treating this guidance as settled unless Anthropic publishes a new policy change.

What Changed

This isn't a reversal — it's clarification. OpenClaw has supported Anthropic's Claude CLI backend for a while, but there was always lingering doubt about whether this violated Anthropic's terms. The key shift: Anthropic staff directly told OpenClaw that this usage is "allowed again," suggesting there may have been a period where it wasn't. The timing matters — this confirmation comes as OpenClaw and similar agent frameworks push harder into CLI-based integrations for their gateways.

The Technical Details

OpenClaw's Claude CLI integration supports both direct "claude -p" usage and the broader CLI reuse pattern. The setup lives in /gateway/cli-backends, and OpenClaw now prefers Claude CLI reuse over the older setup-token approach when available. That said, Anthropic's public Claude Code docs still document "claude -p" as a supported feature — this confirmation just aligns OpenClaw's implementation with explicit blessing from Anthropic staff.

Production Considerations

Here's where it gets practical: while CLI usage is allowed, OpenClaw still recommends Anthropic API keys as "the clearest production path for always-on gateway hosts and explicit server-side billing control." The CLI approach works, but API keys give you more predictable cost visibility and fewer moving parts in production. If you're running agent infrastructure at scale, the API path remains the responsible choice — even if Claude CLI reuse now has the thumbs-up from Anthropic.

Key Takeaways

  • Anthropic staff explicitly confirmed OpenClaw-style Claude CLI usage is sanctioned, marking a clear shift from earlier uncertainty
  • The "claude -p" pattern and CLI reuse are now treated as settled policy unless Anthropic publishes new restrictions
  • API keys remain the recommended production path for billing clarity and always-on gateways
  • This affects OpenClaw users specifically, but signals broader acceptance of CLI-based agent integrations

The Bottom Line

This is a win for OpenClaw and anyone building CLI-based agent workflows, but don't mistake "allowed" for "optimized." Anthropic gave the green light, but for production systems where you need predictable billing and uptime, stick with API keys. The CLI path works — it's just not the enterprise-grade choice.