Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Binary file added public/blog/mcp-new-standard/images/cover.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added public/blog/mcp-new-standard/images/og.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
39 changes: 39 additions & 0 deletions src/content/blog/mcp-new-standard/article.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,39 @@
---
title: "MCP's New Standard: What Changed and Why It Was Needed"
subtitle: 'Stateless core, MCP Apps, one foundation: the July 28 release explained by someone running two production MCP endpoints.'
description: 'MCP went stateless on July 28. A production MCP operator explains what changed, why it was needed, and measures the migration across 8,588 live servers.'
date: 2026-08-04
category: coding
cover: images/cover.png
readingTime: '8 min read'
mediumUrl: 'https://medium.com/@vreshch/mcps-new-standard-what-changed-and-why-it-was-needed-8ac7556183e7'
tags:
- mcp
- agents
- software-architecture
- agentage
- developer-tooling
---

MCP went stateless on July 28, 2026. My server at [memory.agentage.io/mcp](https://agentage.io) had been running that way since May: one server instance per request, no session table, nothing to lose when an instance dies. The first commit of that transport is dated eight days after the release candidate locked. I did not predict anything - I read the roadmap and believed it.

Five days after the spec went final, I probed every remote endpoint in the official registry - 8,588 URLs - to measure whether the migration is real. It is, and its shape says something about how modern standards actually land.

_The request survives. The servers that answered it do not remember._

## What the article covers

1. **Why the handshake had to go.** The session model was honest on a laptop and a trap on a fleet: sticky routing, shared session stores, sessions dying with their instance.
2. **The stateless future was announced in December.** The most telegraphed change in the protocol's history - direction in December, proposal in April, candidate in May, final in July.
3. **The week-one numbers.** My directory at [catalog.agentage.io](https://catalog.agentage.io) indexes 19,598 servers, half of them remote. Of the probed fleet, 293 already negotiate the new spec - and 76.5 percent behave statelessly regardless of what their version field says.
4. **What July 28 does not fix.** The state moved rather than disappeared, deprecations start a twelve-month clock, and context bloat is untouched.

> _Adoption of a well-signposted standard is not an event. It is a version field catching up to behavior._

## Where MCP goes from here

MCP now lives under the Agentic AI Foundation at the Linux Foundation, its SDKs have passed a billion downloads, and GitHub and Cloudflare shipped support before the spec was even final. The argument about whether to adopt it is quietly over; the argument about how to run it well is just beginning. The full article walks through both - with the probe methodology, the operator checklist, and the receipts.

---

Related: [catalog.agentage.io](https://catalog.agentage.io/mcp) - the MCP directory behind the numbers · [agentage Memory](https://agentage.io) - the production endpoint in the story · [@vreshch on GitHub](https://github.com/vreshch).