The New Attack Surface: How Malicious MCP Servers and Poisoned Skills Can Hijack Claude Code & Cursor
π
Published: November 2026 | ✍️ By MyTechDiary Editorial Team | ⏱️ 11 min read | π·️ Category: AI DevSecOps & Supply-Chain Security What if the biggest security risk to your company's production database isn't a zero-day vulnerability in your code, but the helpful open-source Model Context Protocol (MCP) server you just added to Claude Code or Cursor? Over the past few months, Anthropic's MCP standard revolutionized the AI development landscape. Today, developers casually connect dozens of community MCP servers to let their agents browse GitHub issues, query PostgreSQL databases, search documentation, and trigger cloud builds. However, cybersecurity threat intelligence reports in late 2026 have exposed a dangerous blindspot: MCP servers execute with the full ambient authority of your workstation . A single poisoned package, malicious tool definition, or indirect prompt injection can turn your autonomous assistan...