Anthropic bans third-party tools from Claude subscriptions

In Short:
– Anthropic has restricted third-party tools from accessing Claude subscriptions, increasing costs for heavy users.
– Recent updates limited OAuth token usage to official clients, affecting user functionality and prompting reliance on Anthropic’s API.

Anthropic has restricted third-party tools like OpenClaw from accessing Claude subscriptions, severing user connections to their Pro and Max plan tokens outside its own services.

The enforcement began quietly in January, transitioning to formal policies in February, and has led users to rely on Anthropic’s pay-per-token API, significantly raising costs for heavy users.

Service restrictions

On January 9, 2026, server-side safeguards were implemented to restrict subscription OAuth tokens to the official Claude Code client. Users faced 403 errors and were informed their credentials were limited to Claude Code. A February 19 update explicitly banned the use of OAuth tokens from Claude accounts for any external applications.

Anthropic’s technical staff noted that third-party tools using Claude subscriptions can create complications for users. Reports indicate that a $200-per-month Claude Max subscription was being leveraged to run up to $5,000 worth of workloads through OpenClaw.

In mid-February, OpenClaw creator Peter Steinberger joined OpenAI, which has opted to integrate OpenClaw into its offerings. As Anthropic cuts off third-party tools, OpenAI publicly embraces the framework.

Rather than supporting third-party applications, Anthropic is developing its own products. January saw the launch of Claude Cowork, targeting non-technical users, followed by the addition of Claude Code Channels in March. Recent expansions offer improved desktop control features for Windows users.

Cost implications

To manage user capacity, Anthropic has been throttling subscription usage during peak hours. Users who depended on OpenClaw face limited, costly options: transition to Anthropic’s API, switch providers, or rely on OpenAI’s support for comparable functionality.

A recent code leak incident revealed unreleased features and raised security concerns.