Critical Vulnerability in MCP Protocol Exposes 200,000 AI Servers to Remote Code Execution

A report by OX Security has identified a critical vulnerability in the Model Context Protocol (MCP) developed by Anthropic, potentially exposing over 200,000 AI servers to remote code execution. The flaw lies in the STDIO interface of the MCP SDK, which executes commands without proper validation, even if the server fails to start. This design decision affects all 11 programming languages supported by MCP, impacting over 150 million downloads and thousands of public servers. Researchers have demonstrated that attackers can exploit this vulnerability to gain complete control over affected systems. Despite multiple disclosures to Anthropic, the company maintains that the behavior is by design and has not implemented any architectural changes. The vulnerability has been assigned a CVE number, and OX Security has documented over 10 high or critical-severity CVEs related to this issue. The situation poses a significant risk to developers and organizations relying on MCP for AI applications.

Key Points: • A critical vulnerability in MCP could affect over 200,000 AI servers. • The flaw allows arbitrary command execution via the STDIO interface without validation. • Anthropic has declined to patch the vulnerability, citing it as expected behavior.