MCP Was Wrong From The Start (They Just Fixed It)

ℹ️

MCP's 2026-07-28 spec makes the entire protocol stateless, deleting the initialize handshake and the Mcp-Session-Id header that pinned clients to one server. That fixes round robin load balancing, removes the Redis instance holding your sessions, and lets a server scale to zero on Workers or Cloud Run. It also brings new Mcp-Method and Mcp-Name HTTP headers, ttlMs and cacheScope caching hints, elicitation rebuilt as Multi Round-Trip Requests, and Tasks graduating from experimental into an official extension. 🔗 Relevant Links Release blog: https://blog.modelcontextprotocol.io/posts/2026-07-28-release-candidate/ ❤️ More about us Radically better observability stack: https://betterstack.com/ Written tutorials: https://betterstack.com/community/ Example projects: https://github.com/BetterStackHQ 📱 Socials Twitter: https://twitter.com/betterstackhq Instagram: https://www.instagram.com/betterstackhq/ TikTok: https://www.tiktok.com/@betterstack LinkedIn: https://www.linkedin.com/company/betterstack 📌 Chapters: 0 - Intro 0 - Why Stateful MCP Broke 1 - What The New Spec Removes 1 - Round Robin And Scale To Zero 2 - HTTP Headers And Cache Hints 3 - Where The Handshake Info Went 3 - Managing State Yourself 4 - Multi Round-Trip Requests 5 - Long Running Tasks 5 - Deprecations And Upgrading 6 - Final Thoughts

Watch or listen at the source

Unknown directive: transcript

Debug Info

json

{
  "name": "transcript",
  "attributes": {}
}