MCP has a BIG problem

MCP has several challenges but the biggest one is the fact that it is a *stateful* protocol. This effectively means that you cannot add MCP compatibility to a simple REST API - you need to deploy a stateful long-running server. This also means that serverless functions - which are a natural fit for AI tool calls - cannot be used with MCP. I believe this is a show-stopper for MCP and that other simpler methods are the best way forward for solving the AI data integration problem. Ultimately, we need to hook up *existing* APIs to Agents - not force developers to create and deploy new types of servers. 📚 Resources: - My previous MCP video: https://youtu.be/m46tZX6vceI - My written thoughts on MCP state: https://github.com/modelcontextprotocol/specification/discussions/102#discussioncomment-12445025 - MCP state full discussion: https://github.com/modelcontextprotocol/specification/discussions/102 - MCP sampling: https://modelcontextprotocol.io/docs/concepts/sampling - Wild-card Agents json: https://docs.wild-card.ai/agentsjson/schema#top-level-structure 🚀 In This Video, You'll learn: - What are MCPs - The biggest issue with MCP - MCP alternatives - What is Agents.json - OpenAPI vs MCP - Agents.json vs MCP - MCPs for beginners - Limitations of MCP - Can MCP be used in serverless functions 💡 Perfect for Viewers Interested in: - AI data integration - AI agent workflows - Building MCPs - Software Development 2025 - Using AI to code - Coding with AI - Latest AI tutorials 🔴 Subscribe for more tutorials on AI and programming! Chapters: 00:00 - The Big Problem with MCP 03:00 - Why does MCP need state? 06:31 - MCP Alternatives 07:44 - A thought on Agent Tool Routing