Loading video player...
Den Delimarsky is the lead maintainer of the Model Context Protocol (MCP) at Anthropic — the open-source protocol that OpenAI, Microsoft, JetBrains, Cursor, and the entire AI industry has adopted in just one year. In this conversation, Den pulls back the curtain on how MCP actually gets built, why most proposals get rejected, and the #1 mistake developers make when building MCP servers. Topics Covered: MCP governance — keeping the protocol simple, the BDFL model, why they reject bloated proposals Enterprise adoption — OpenAI, Microsoft, JetBrains, Cursor all using MCP; enterprises running internal MCP servers in production The #1 mistake building MCP servers — mirroring your entire API instead of abstracting around customer scenarios MCP Tasks — new async/long-running process support (e.g., video rendering without blocking the client) Agent-to-agent communication — MCP vs A2A, text file state sharing, the future MCP Apps — rendering interactive UIs inside chat interfaces Skills & brainstorming — using Claude Code skills to bootstrap MCP servers, the /clarify pattern Context bloat — dynamic tool discovery and filtering as the solution Who should build MCP servers — same engineering teams that build your APIs Get involved in the Agentic AI Foundation: aaif.io Demetrios: https://www.linkedin.com/in/dpbrinkm/ Den: https://www.linkedin.com/in/dendeli/