← Publications

MCP-SecLint: An Open-Source Static Analyzer for Detecting Vulnerabilities in LLM Tool Integrations

AP. Ridao, M. Safari, Z. Kang & N. Dragoni

Abstract

As the Model Context Protocol (MCP) emerges as a leading standard for enabling large language models (LLMs) to interact with external tools and services, it also introduces new security concerns. Although recent work has identified MCP-specific attack vectors, automated detection tools remain limited. We present MCP-SecLint (MCP Security Linter)—an open-source static analysis tool that detects vulnerabilities in the source code of MCP server implementations. MCP-SecLint targets MCP servers implemented in JavaScript and TypeScript and works by tracking the flow of untrusted tool arguments and sensitive data to dangerous sinks. The system comprises three core analyzers: (1) detection of dangerous command execution, (2) identification of token passthrough vulnerabilities, and (3) detection of unauthenticated endpoints, integrated within a modular, extensible architecture. We evaluate MCP-SecLint by using it on 100 publicly available MCP server repositories found on GitHub, revealing a 5% vulnerability rate. These results demonstrate the prevalence of security weaknesses in the current MCP ecosystem, underscoring the need for protocol-aware security tooling that can detect vulnerable code early in the development lifecycle.
MCP-SecLint architecture: JavaScript and TypeScript source flows through an AST parser, three vulnerability analyzers, and SARIF, JSON, or console output formatters
Open original