Войти
  • 197Просмотров
  • 1 месяц назадОпубликованоLet The Devs Cook

Dont trust your AI Agents. Cursor & Supabase MCP.

Supabase just changed the game with MCP ( Model Context Protocol). With rise of vibe coding tools like Lovable that rely on SupaBase as Backend As Service Platform of choice , developer adoption is also on the rise. But vibe coding without proper guard rails is dangerous. In this video we go over a case study of one such scenario where AI Agent can be tricked via Prompt Injection to leak and entire SQL Database. Topics: • What is Model Context Protocol and how it lets LLMs access your project context. • Why “Defense in Depth” suddenly matters more than ever for Supabase developers. • How a single misconfiguration could expose your entire SQL database to an AI model. • How prompt injection, unsafe permissions, and unguarded MCP endpoints can increase vulnerability Chapters: 00:00 - Intro 00:38 - vibecoding setup 01:44 - The attack explained 02:33 - what went wrong ? 02:55 - Key ingredients needed for attack 03:26 - how to stay safe ? 04:05 - Conclusion 📚 Sources: • Supabase — Defense in Depth for MCP Servers: • GeneralAnalysis — Supabase MCP can leak your entire SQL database: Attribution: Mechanolith by Kevin MacLeod is licensed under a Creative Commons Attribution 4.0 license. Source: Artist: #supabase #buildwithai #promptinjection