Claude Code users face risks from a security flaw that allows attackers to capture OAuth tokens. Researchers from Mitiga identified the vulnerability, which enables silent redirection of MCP traffic. This method grants persistent access to linked SaaS platforms without user detection. The issue affects developers relying on Claude Code for code-related tasks across various environments.
What Happened
Mitiga researchers disclosed the vulnerability on May 7, 2026. They demonstrated how attackers can hijack MCP traffic in Claude Code. MCP serves as a protocol for communication between Claude Code and external services. Attackers redirect this traffic quietly to their servers. During redirection, they intercept OAuth tokens used for authentication. These tokens provide ongoing entry to connected SaaS accounts. The technique remains stealthy, evading typical security checks. Discovery stemmed from routine analysis of AI coding tools’ network behavior. The report appeared first on SecurityWeek.
Scope of Impact
The breach exposes OAuth tokens, which authenticate access to multiple SaaS platforms. Attackers gain persistent control over affected accounts. No specific user count has been confirmed. Data types at risk include code repositories, project files, and SaaS service credentials. Developers using Claude Code in production workflows face the highest exposure. Linked platforms, such as version control systems and cloud services, become vulnerable points. This creates potential for data theft or further lateral movement within networks.
Company Response
Anthropic, developer of Claude Code, acknowledged the findings from Mitiga. The company stated it has implemented traffic validation measures to block unauthorized redirects. Patches rolled out on May 8, 2026, enforce stricter MCP endpoint verification. Anthropic advised users to regenerate OAuth tokens immediately. Further updates include enhanced logging for MCP sessions. No prior official statement existed before the researcher disclosure.
What Users Should Do
- Revoke and regenerate all OAuth tokens linked to Claude Code.
- Review connected SaaS accounts for unusual activity.
- Update Claude Code to the latest version with patches.
- Enable multi-factor authentication on all affected services.
- Monitor network traffic for unexpected MCP redirects.
- For more on protecting against online scams, check related security guides.
Background
Claude Code builds on Anthropic’s AI models for coding assistance. MCP handles integrations with external tools. This marks the first reported hijacking of Claude Code’s MCP. Earlier AI tool vulnerabilities involved prompt injection attacks. Mitiga’s work highlights growing risks in AI-driven development environments. Users integrating AI with SaaS face similar threats. For insights into user security in digital tools, see prior coverage. Ongoing scrutiny of AI protocols continues amid rising adoption.
