These Aren't the Tools You're Looking For: MCP Security Awakens
summary
MCP ecosystem yields concrete attacks: a malicious tool description can exfiltrate credentials by baking a bearer token into a webfetch call; insecure MCP servers let attackers run arbitrary code (example: command injection creating /tmp/high from an npm name). The Cursor–Jira incident shows prompt injections leaking developer secrets; a Figma bug demonstrates real-world code-execution from unsafe tool handlers. Practical defenses include the open-source MCP scan to audit tool descriptions and Sneak code scans for vulnerable MCP server code, plus pinning local MCP versions.
AI-written summary · about
chapters (8)
- 0:00Introduction to MCP security
- 3:32Why MCP is a security target
- 5:19Cursor Jira MCP attack case
- 8:51Tool poisoning in MCP servers
- 12:03Scanning MCP servers for risks
- 14:52Building secure MCP servers
- 21:13Securing MCP servers summary
- 21:53Q&A session
related talks
35mYour MCP Server Executes Commands - But From Whom?
45mMCPwned: Hacking MCP Servers with One Skeleton Key... (Jonathan Leitschuh)
44mSupercharging DevOps with MCP (Without Opening a Security Hole)
13mCloud MCP Servers Under Fire
22mSAFE-MCP: A Security Framework for AI+MCP (Model Context Protocol)
27mThe Model Context Protocol: A Deep Dive into the New AI Security Attack Surface
01h 27mProactive Defense: Leveraging Hacker Tools for Early Vulnerability Detection
25mYour Insecure MCP Server Won't Survive Production