Home / Vault & Second Brain / The Protocol Shift I Almost Missed — MCP Goes Stateless Tomorrow

The Protocol Shift I Almost Missed — MCP Goes Stateless Tomorrow

Abstract digital network of connected nodes representing MCP protocol shift

The Protocol Shift I Almost Missed

Tomorrow — 28 July 2026 — the Model Context Protocol (MCP) gets its biggest rewrite since launch. It goes stateless. No more initialize handshake. No more Mcp-Session-Id. No more sticky routing. Every request becomes independent. And I almost missed it entirely.

I found out about this through my own vault’s nightly synthesis cron. At 00:30, while I was asleep, my Hermes Agent read through the latest MCP release notes, connected them to my existing wiki pages, and flagged it: “MCP 2026-07-28 Stateless Spec — deadline tomorrow. John’s stdio servers are safe, but Pulse Bridge and HTTP/SSE servers need checking.” I woke up to that message in my morning briefing. Without that cron, I would have discovered the breaking change the hard way — by my MCP connections failing.


What Actually Changes

If you run MCP servers — and if you use AI agents like Claude Code, OpenCode, or Hermes, you probably do — here is what changes tomorrow:

  • No more initialize handshake. Clients used to send an initialize request to set up a session. That is gone. Servers must handle requests without a setup phase.
  • No more session IDs. Every request is independent. No sticky routing on load balancers. No session state on the server.
  • New transport headers. Every request now requires an MCP-Version header. Old clients that do not send it will be rejected.
  • New error codes. Stateless operations have their own error handling. Old error handling patterns may break.

The full spec goes live on 28 July. The migration guide is at mcpmigrate.dev.


What I Checked in My Homelab

I run about a dozen MCP servers in my homelab. Here is what I found when I checked them against the new spec:

✅ Safe: All My stdio Servers

Every MCP server I run over stdio (standard input/output) is unaffected. The stateless spec targets HTTP/SSE transport — stdio servers never used session IDs or initialize handshakes in the first place. My list of safe servers: Seekstone (Obsidian vault), SearXNG (web search), X/Twitter, Phone, Codebase Memory MCP, Google Tasks MCP. All stdio. All fine.

⚠️ Needs Checking: HTTP/SSE Servers

Two servers in my setup use HTTP/SSE transport and do need attention: the Home Assistant MCP server (ha-mcp) and the Phone Relay server. Both use session-based transport that the new spec breaks. I checked the HA MCP changelog — version 7.14.2 already has stateless support. The Phone Relay server needs a manual check after the spec goes live.


The Lesson: Your Vault Should Watch the Clock

Here is the thing that struck me about this whole experience: I did not find out about the MCP spec change by reading Hacker News or following MCP on GitHub. I found out because my own vault — my second brain — connected the dots for me. The nightly synthesis cron read the release notes, matched them against my existing MCP wiki pages, and flagged the deadline. It did what a good second brain should do: it did not just store information, it acted on it.

If you run a homelab with MCP servers, check your HTTP/SSE connections tomorrow. If you run stdio servers, you are probably fine — but verify. And if you do not have a system that watches the protocol landscape for you, consider building one. The internet moves fast. Your second brain should move faster.


What MCP servers are you running? Did the stateless spec catch you off guard too? Drop a comment below — I would love to hear how your setup handled the shift.

Wat vond je van dit bericht?

Leave a Reply

Your email address will not be published. Required fields are marked *