Anthropic Restricts Third-Party API Access
Anthropic has halted third-party OAuth access for its Claude subscriptions, impacting developers and tools that rely on integrations. The move is intended to prevent IP theft and encourage use of its paid API, but has sparked debate in the developer community about API stability and platform risk.
- The technical enforcement began with blocks against third-party clients between January 9 and 27, 2026, with a formal documentation update clarifying the policy on February 19, 2026. This move disrupted developers who had built workflows and tools on top of Claude subscriptions without prior notice. - Anthropic's stated reasons for the change include preventing the use of third-party tools to "spoof" their official Claude Code harness, which generated unusual traffic and made it difficult for them to debug user issues. The company aims to direct developers to its paid API, ensuring proper billing and control over its intellectual property. - This policy change forces developers building products on Claude to use official API keys, shifting from a flat-rate subscription model to a usage-based pricing model, which can significantly increase costs for heavy users. This has been described as a move to prevent "token arbitrage," where users accessed models via cheaper subscriptions for work that would have been more expensive through the API. - The change significantly impacted numerous third-party tools, with OpenCode, a popular open-source coding agent with over 107,000 GitHub stars, being a primary casualty. Other affected tools include Cline, Roo Code, and OpenClaw. - This action is not an isolated incident; Anthropic has previously restricted access to its models for other companies. In June 2025, it cut most of Windsurf's direct access with little notice, and in August 2025, it revoked OpenAI's access, citing a terms of service violation. - The developer community has raised concerns about "platform risk," where a foundational platform can unilaterally make changes that break dependent applications and businesses. This has led to discussions about the importance of diversifying AI model dependencies and favoring open-source models to avoid such risks. - An Anthropic engineer, Thariq Shihipar, acknowledged that the documentation update was confusing and advised users building businesses on their Agent SDK to use API keys. This highlights the tension between encouraging experimentation and enforcing commercial terms. - The move is seen by some as a broader industry trend where AI companies are tightening control over their ecosystems to ensure economic sustainability and prevent the disintermediation of their models by third-party tools that offer a better user experience.
