Researchers with Cyata and BlueRock uncovered vulnerabilities in MCP servers from Anthropic and Microsoft, feeding ongoing security worries about MCP and other agentic AI tools and their dual natures ...