Veylith

Veylith is an open-source, research-alpha security gateway for AI agents using the Model Context Protocol (MCP). As AI agents interact with external tools, malicious tool descriptions, unsafe arguments, and indirect prompt injections can expose them to security risks. Veylith introduces an inline security boundary between AI clients and MCP servers, inspecting tool discovery, invocation arguments, and returned content. Built in TypeScript, Veylith combines deterministic security rules, heuristic screening, optional semantic model evaluation, and tamper-evident SHA-256 audit logging. It provides an MCP gateway, stdio integration, CLI, research dashboard, and automated testing. The project demonstrates how security enforcement can become part of the infrastructure surrounding AI agents. Veylith remains an early research prototype; its evaluation results are limited to documented test scenarios and do not represent a universal security guarantee.