Claude Subscriptions Will No Longer Cover Third-Party Tool Usage

Boris Cherny, the creator of Claude Code, posted the news today: starting tomorrow (April 4) at 12pm PT, your Claude subscription no longer covers usage on third-party tools.

If you’re running OpenClaw, NanoClaw, or any other third-party harness that authenticates through your Claude login, you have two options going forward. You can buy extra usage bundles (which Anthropic says are now available at a discount), or you can switch to a Claude API key and pay per token.

Your subscription still works for Claude Code and claude.ai. Nothing changes there.

How We Got Here

This has been building for months. Here’s the timeline:

In late 2025, developers figured out they could extract the OAuth token from a Claude Pro or Max subscription and feed it into third-party tools like OpenClaw. A $200/month Max subscription got you roughly the same model access that would cost $1,000 or more through the API. The math was obvious, and the workaround spread fast.

On January 9, 2026, Anthropic flipped a switch. Server-side changes blocked subscription OAuth tokens from working outside the official Claude Code client. No warning, no grace period. Developers woke up to broken workflows.

In February, Anthropic updated its Consumer Terms of Service to make the policy explicit: OAuth tokens from Free, Pro, and Max accounts are “intended exclusively for Claude Code and Claude.ai.” Using them anywhere else is a violation.

The reaction was not warm. DHH called the move “very customer hostile.” The Hacker News thread pulled 245+ upvotes. A GitHub issue on the topic got 147+ reactions.

Now, with today’s announcement, Anthropic is drawing a clean line with a specific date instead of just quietly breaking things.

What “Extra Usage Bundles” Means

Anthropic’s subscription plans (Pro at $20/month, Max at $100 or $200/month) give you a fixed allotment of usage on Claude Code and claude.ai. Extra usage bundles let you buy additional capacity on top of that allotment, billed at API-adjacent rates with a user-controlled monthly spending cap.

Cherny’s tweet says these bundles are “now available at a discount,” though he didn’t specify the discount amount. What we do know is that typical API usage for developers runs about $12/day, which works out to roughly $360/month. Prompt caching and batch API discounts can cut that significantly.

The framing here matters. Anthropic isn’t saying “stop using third-party tools.” They’re saying “stop using your flat-rate subscription to fund unlimited third-party tool usage.” If you want to keep using OpenClaw with your Claude login, you can, but you’ll pay for what you use through the bundles.

Why Anthropic Is Doing This

Three factors are at play.

Revenue math. The subscription arbitrage was real. A $200/month Max plan gave heavy users access to thousands of dollars in compute. That pricing gap was never sustainable, and Anthropic’s investors (who just put in $30 billion at a $380 billion valuation) presumably want the economics to work.

Safety and telemetry. Anthropic has said that third-party harnesses bypass the telemetry and safety guardrails built into Claude Code. The Augmented Mind breakdown noted that Anthropic identified patterns suggesting “influence operations and intrusion workflows” running through spoofed Claude Code clients. Whether that’s the primary motivation or a convenient justification depends on your level of cynicism.

Product strategy. Anthropic has been building its own alternatives to the features that made OpenClaw attractive. Claude Code Channels shipped in March, letting you control Claude Code through Telegram and Discord. Remote Control lets you hand off sessions to your phone. Scheduled tasks handle background automation. The gap between what OpenClaw offers and what Claude Code offers natively has been shrinking.

What You Should Do

If you’re currently running third-party tools on your Claude subscription, here’s how to think about the transition:

Check your usage patterns. If you’re a light user who runs OpenClaw for occasional tasks, the extra usage bundles with the new discount might cost less than you’d expect. If you’re a heavy user running automated agents all day, API keys with prompt caching are probably cheaper.

Consider the API directly. Claude API pricing for Sonnet-class models runs $3-$15 per million tokens. With prompt caching (up to 90% savings on cached content) and the batch API (50% discount), sustained usage can land in the $100-$200/month range for many workflows.

Look at alternatives. Some developers have already moved to multi-model setups. One widely shared approach runs OpenClaw on Kimi K2.5 for $15/month. Others use a mix of local models for bulk work and Claude API for tasks that need top-tier reasoning.

Keep your architecture flexible. The developers who got burned worst in January were the ones locked into a single model provider with a single authentication method. Building for model portability, where you can swap providers without rewriting your stack, protects you from the next policy change, whoever it comes from.

The Bigger Picture

Every major AI lab is drawing the same line that Anthropic just drew. Fixed-price subscriptions are for the vendor’s own products. Third-party access goes through metered APIs. OpenAI did this with ChatGPT and Codex. Google does it with Gemini. The era of using a consumer subscription as an all-you-can-eat API proxy is ending across the board.

For developers, the practical takeaway is straightforward: subscriptions are for interactive use in official tools, APIs are for building and integrating. The sooner your setup reflects that distinction, the less disruption you’ll face next time a vendor tightens the rules.