Threat Modeling the Model Context Protocol: Securing Agentic Tools with mcpscan

The Model Context Protocol (MCP) is revolutionizing how LLMs interact with local and remote environments by enabling active agentic execution. However, this bridge between AI reasoning and system-level tools introduces significant security risks, including indirect prompt injection, command injection, and credential exfiltration. To address these vulnerabilities, the open-source tool 'mcpscan' has been introduced. It provides static analysis capabilities to audit MCP server implementations and local configurations for common security flaws. By scanning for unsafe subprocess calls, hardcoded secrets, and over-privileged directory access, mcpscan helps developers secure their agentic workflows. The tool supports SARIF output for seamless integration into CI/CD pipelines, such as GitHub Actions. This article emphasizes the importance of defense-in-depth practices, such as strict workspace scoping and sanitizing inputs, to ensure that the integration of AI agents into development environments remains secure against emerging attack vectors.