Anthropic Pushes Third-Party Tools, Especially OpenClaw, Behind a Paywall

AI provider Anthropic is gradually reshaping its relationship with third-party tools such as OpenClaw. While external applications will in future have to draw from a paid credit balance, Anthropic is expanding its own chatbot Claude with precisely the capabilities for which OpenClaw has so far been known. The development raises questions about how independent third-party providers can survive long-term within the Claude ecosystem.

Since 4 April 2025, an important change has applied to all users who have connected external applications such as OpenClaw to their Claude account. Such so-called third-party harnesses no longer draw on the usage quota included in the subscription, but instead draw exclusively from the separate "Extra Usage" credit balance. This means: anyone using OpenClaw or similar tools must pay additionally, even if the regular monthly budget has not yet been exhausted.

Anthropic communicates the change matter-of-factly and without explanation. For users who have until now relied on third-party integrations at no extra cost within their subscription, this represents a noticeable additional burden. The rule applies to Pro, Max, and Team plans.

To help users offset the increased costs, Anthropic offers so-called Usage Bundles. These credit packages can be purchased in advance and apply to all Claude products -- that is, Claude Web, Desktop, Mobile, Claude Code, Cowork, and also third-party tools. Larger packages come with higher discounts.

Pro and Max users can purchase packages with a combined value of up to $2,000 per month at a discount. For Team plans, the limit is $3,000 per month. Usage beyond these caps is billed at standard prices without a discount.

In parallel with the decoupling of external tools, Anthropic is expanding Claude with features that have until now been the unique selling point of OpenClaw and similar third-party providers. With the new "Computer Use" feature, Claude can directly control a user's screen -- clicking, typing, and opening applications, just like a human user. The feature is currently available exclusively to paying users on Pro and Max plans and runs on macOS.

Technically, Claude works with continuous screenshots to recognise the current state of the screen and respond accordingly. The AI prioritises direct API connections to services such as Gmail or Slack, falls back on the browser when no such connection exists, and only takes over direct mouse and keyboard control as a last resort.

"Computer Use is a new capability, and the threats it is designed to protect against are constantly evolving. Claude makes mistakes, and no safeguards are perfect."

The fact that Anthropic acquired the startup Vercept in February 2026 to further develop "Computer Use" underscores this strategic direction. Anthropic is deliberately investing in the very capabilities that external providers such as OpenClaw have until now brought into the Claude ecosystem. Observers see a clear pattern in this: third-party providers are being financially disadvantaged while Anthropic integrates their core features into its own product.

Despite the expansion of features, Anthropic emphasises that "Computer Use" still has significant limitations. The feature is currently only available on macOS; Windows support is planned. Complex multi-step tasks can fail, and screen control is considerably slower than direct API connections. Team and Enterprise plans do not have access for the time being.

On the security side, Claude requests explicit user approval before accessing any application. Investment and crypto platforms are blocked by default. Since Claude continuously takes screenshots for screen control, Anthropic explicitly recommends closing sensitive applications before use.