MCP Security: Vetting Servers to Mitigate Tool Poisoning Attacks

This video reveals my systematic approach to evaluating MCP servers before installation to protect security and privacy. With thousands of MCP servers now available, I show you: My personal criteria for determining which MCP servers are safe to use The Claude Project i made using MCP servers to evaluate other MCP servers A live demo evaluating the Apple MCP server Custom instructions you can copy and use today (GitHub link below) This practical guide helps you navigate the explosion of MCP servers while protecting your sensitive data and system access. Don't install unknown MCP servers without doing some due dilligence! 🔗 Helpful Links: - Need personalized guidance? Book a call: https://cal.com/jeredblu - GitHub Repo with custom instructions: https://github.com/JeredBlu/custom-instructions/blob/main/mcpevaluatorv3.md Setup Videos: - Setting up filesystem: https://youtu.be/7l4vTHYpYUw - Setting up brave search: https://youtu.be/sWjrfJcMWEQ - Setting up sequential thinking: https://youtu.be/R-5ucM-5P5o Timestamps: 00:00 - Introduction to MCP server explosion 00:21 - MCP is now industry standard 00:54 - The challenge of too many MCP servers 01:26 - Security and privacy considerations 01:48 - My manual evaluation methodology 02:42 - Demo of MCP Evaluator project with Claude 04:04 - Step-by-step analysis process 06:30 - Security assessment and results review 07:45 - Dealing with suspicious findings 08:42 - Building your own improved MCP servers 09:34 - Tips for safely using MCP servers 10:51 - Future of MCP server standards 11:54 - Closing thoughts #MCP #TPA #ModelContextProtocol #AISecurity #github #Claude37 #MCPServers #AIPrivacy #CyberSecurity #CloudAI #AIProductivity #invariant