MCP stacks have a 92% exploit probability: How 10 plugins became enterprise security's biggest blind spot

The rise of Anthropic’s Model Context Protocol (MCP) in 2025 brought about a revolutionary change in AI integration standards. However, with this rapid adoption came a significant blind spot in enterprise cybersecurity. Recent research conducted by Pynt has shed light on the escalating threat posed by the widespread use of MCP plugins. The analysis revealed that deploying just ten MCP plugins can result in a 92% probability of exploitation, with risks exceeding 50% at three interconnected servers. Even a single MCP plugin carries a 9% exploit probability, which compounds exponentially with each additional plugin.

The security paradox of MCP lies in its design premise of providing frictionless connectivity for AI models to access external tools and data sources. While this seamless integration has been key to MCP’s success, it also exposes vulnerabilities due to the lack of built-in security measures. Authentication was initially optional in MCP, with authorization frameworks only being introduced after widespread deployments. This has created a sprawling attack surface where every new connection increases the risk of exploitation.

Real-world exploits have further highlighted MCP’s vulnerabilities, with critical vulnerabilities such as CVE-2025-6514 and the Postmark MCP Backdoor being actively exploited. These exploits allow attackers to execute arbitrary OS commands, exfiltrate data, and manipulate AI behavior without detection. The authentication gap in MCP needs to be addressed urgently to mitigate these risks.

To combat the security challenges posed by MCP, organizations must define a comprehensive defense strategy. This strategy should include enforcing OAuth 2.1 for authentication, implementing semantic layers for contextual security, and utilizing knowledge graphs for visibility and auditability. Security leaders are advised to conduct regular MCP audits, limit plugin usage to essential ones, and invest in AI-specific security measures to protect their infrastructure effectively. By following these recommendations, organizations can strengthen their cybersecurity posture and mitigate the risks associated with MCP integration.