MCP Just Deleted Its Own Handshake — and Every Request Got 118% Bigger
Last Updated on July 27, 2026 by Editorial Team
Author(s): Chew Loong Nian – AI ENGINEER
Originally published on Towards AI.
MCP Just Deleted Its Own Handshake — and Every Request Got 118% Bigger
I measured the new Model Context Protocol wire format against the old one this morning, and the same tools/call went from 190 bytes to 414 bytes. That is a 118% increase on every single request, and the break-even point against the handshake it replaces is less than one call.
After MCP’s July 28, 2026 revision removes the initialize handshake and the Mcp-Session-Id header, the article explains what disappears (session negotiation and sticky routing) and what replaces it (per-request _meta, plus server discovery and explicit routing headers). The author then details an experiment showing the one-time handshake cost is small, but repeating _meta on every call makes requests balloon over multiple tool invocations, leading to a large linear overhead at scale. It also covers caveats (bandwidth vs token usage; overhead ceiling depends on which metadata is included), the “handle pattern” as the new way to model state (including security implications of handles appearing in chat transcripts), and additional protocol changes such as deprecated features and deprecation policy. Finally, it advises when to target each protocol version (local stdio vs remote load-balanced servers), summarizes SDK/upgrade guidance, and concludes with a verdict that while the redesign fixes real infrastructure problems, it shifts complexity and security responsibility from the platform layer to individual server authors.
Read the full blog for free on Medium.
Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.
Published via Towards AI
Towards AI Academy
We Build Enterprise-Grade AI. We'll Teach You to Master It Too.
15 engineers. 100,000+ students. Towards AI Academy teaches what actually survives production.
Start free — no commitment:
→ 6-Day Agentic AI Engineering Email Guide — one practical lesson per day
→ Agents Architecture Cheatsheet — 3 years of architecture decisions in 6 pages
Our courses:
→ AI Engineering Certification — 90+ lessons from project selection to deployed product. The most comprehensive practical LLM course out there.
→ Agent Engineering Course — Hands on with production agent architectures, memory, routing, and eval frameworks — built from real enterprise engagements.
→ AI for Work — Understand, evaluate, and apply AI for complex work tasks.
Note: Article content contains the views of the contributing authors and not Towards AI.
Recent Posts
