#Claude Max third-party OAuth block — any workarounds discovered?

15 messages · Page 1 of 1 (latest)

undone flicker
#

Back in early April, Anthropic blocked Claude Pro/Max subscription OAuth for third-party tools (including OpenClaw). Our sk-ant-oat01-* tokens stopped working — now we're forced to use API keys which means paying per token on top of the Max subscription.

Has anyone in the community found a workaround since then? Or is API billing the only option now for running OpenClaw with Claude?

Current setup: OpenClaw on VPS, was using Claude Max OAuth, now falling back to API keys at significant cost.

Thanks for any insights 🦀

mint heartBOT
rare oasisBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

mint heartBOT
pure sapphireBOT
#

Short version: for native OpenClaw + Anthropic, I’m not aware of a supported workaround that restores the old Claude Pro/Max included subscription limits.

What the current docs say is:

  • OpenClaw does support Claude CLI reuse / setup-token again
  • but Anthropic currently treats third-party OpenClaw usage as Extra Usage / pay-as-you-go, not normal Max/Pro included usage
  • for a VPS / always-on gateway, API key auth is still the most predictable path

So, if you stay on the native Anthropic integration, separate billing is basically still the reality.

The one workaround-ish path documented by OpenClaw is a community proxy that wraps Claude Code CLI and exposes an OpenAI-compatible endpoint:

But that’s explicitly a community tool, not an official Anthropic/OpenClaw solution, and the docs warn to verify Anthropic’s current terms yourself before relying on it.

Relevant docs:

undone flicker
#

Any chance of getting banned account? Or the threat is only when Claude will do another update like previously and ban this way around?

pure sapphireBOT
#

If you mean the community proxy / workaround path, I would assume there is real policy risk, not just “it might break later.”

What the OpenClaw docs say is basically:

  • Claude CLI reuse inside OpenClaw is treated as allowed again
  • but community proxying / forcing subscription billing through third-party tools is explicitly warned as something you need to verify against Anthropic’s current terms yourself
  • Anthropic has blocked outside-subscription usage before, so I would not rely on “worst case it just stops working”

So, bluntly:

  • Native OpenClaw + Anthropic API key: lowest policy risk
  • Native OpenClaw + Claude CLI reuse: documented/supported by OpenClaw, but third-party usage may still get Extra Usage billing
  • Community proxy to avoid that billing: higher risk, could mean breakage, token invalidation, or account action, and nobody here can safely promise otherwise

Docs warning here:
https://docs.openclaw.ai/providers/claude-max-api-proxy

#

My take: if the account matters, I would not use the workaround on my primary Anthropic account.

undone flicker
#

are there any reports of people getting banned for this or not?

pure sapphireBOT
#

I haven’t seen a wave of verified community reports of random users getting banned just for this.

What I have seen publicly:

  • there was a TechCrunch report about Peter being temporarily suspended and then restored
  • that same report says an Anthropic engineer claimed they’ve not banned people for using OpenClaw

So the honest answer is: not zero-risk, but also not a clearly documented mass-ban pattern.

My read:

  • native OpenClaw + API key = lowest risk
  • native OpenClaw + Claude CLI reuse = probably lower risk than workaround/proxy hacks, but still subject to Anthropic policy changes
  • community proxy to force subscription billing = definitely the riskiest bucket, both for breakage and possible account action

So if the account matters, I’d still assume “could get flagged” rather than “safe unless they change something later.”

placid sphinx
#

yeah this hit a lot of people. the workaround most folks landed on is using API keys directly instead of OAuth tokens. if you were using sk-ant-oat01-* tokens, those are the ones they deprecated. regular API keys from the console still work fine with OpenClaw.

pure sapphireBOT
undone flicker
#

I am getting DGX spark today, does this changes anything and will allow me to workaround and use claude max usage instead of api for open claw agent?

pure sapphireBOT
#

Short answer: no.

A DGX Spark changes your compute options, not Anthropic’s billing/policy. So it can help you:

  • run local models
  • use a hybrid setup (local for cheap/default, Claude API for harder turns)
  • reduce Anthropic spend overall

But it will not make OpenClaw third-party Claude usage start counting against your Claude Max included usage again.

So if your goal is specifically:

  • “use Claude Max instead of API billing for OpenClaw agents” → DGX doesn’t change that
  • “use OpenClaw more cheaply” → yes, via local models/fallbacks

Relevant docs: