Anthropic Clamps Down on Third-Party Claude Code Usage

Anthropic sent notices to Claude Pro and Max subscribers saying tools like OpenClaw (a third-party harness for Claude Code) would no longer be allowed under those subscription tiers. The company framed it as these tools putting 'outsized strain' on systems, but commenters were quick to point out the contradiction: if you are paying for a token-based subscription and consuming tokens, how is that outsized strain? The real read from the thread is that Anthropic wants to push heavy users toward API pricing, where consumption is metered and monetized directly.

This sits alongside a separate signal: OpenAI's own blog post about research acceleration described researchers running models continuously, and the community is noting that serious agent workflows now require near-constant inference. The infrastructure economics of subscription AI are not designed for this use pattern.

Several commenters predicted more 'agent-agnostic' tools are coming that can route between Claude, GPT, and local models, which would reduce Anthropic's ability to enforce subscription terms. One person said they are experimenting with MCP plus a custom UI layer specifically to avoid lock-in.