MCP completes its largest rewrite since launch, dropping protocol-level sessions so any server replica can handle any request
The Model Context Protocol has finished its biggest architectural change since its 2024 debut: the new specification removes protocol-level sessions and the 'initialize' handshake, letting requests run behind ordinary load balancers with no sticky routing or shared session store. Any replica can now serve any request — the change that lets agent backends scale like normal web services.
The removed piece is session affinity, and its removal is the whole story. Requiring every message in a conversation to return to the server that holds its state is what made agent fleets expensive, because it defeats the load balancers and autoscalers that make ordinary web traffic cheap. Self-contained, stateless requests can be routed, cached, and traced by any replica — the difference between a bespoke deployment and commodity infrastructure.
The adoption numbers explain why the change matters now rather than as a curiosity. MCP SDK downloads reached roughly 97 million a month by early 2026, and around 41% of software-industry technical leaders report their organisation already runs MCP in some production capacity. A protocol at that scale cannot afford an architecture that fights horizontal scaling.
The governance context is the reassurance underneath. MCP was donated to the Linux Foundation's Agentic AI Foundation in December 2025 with OpenAI, Google, Microsoft, AWS, Salesforce, and Snowflake as backers, and this is the first major spec change under that shared stewardship. It reads like an operator's wishlist, not one vendor's — which is what a standard needs to become infrastructure.
MLQ.ai — MCP's stateless rewrite targets enterprise-scale AI agents → · The Agentics — The enterprise MCP guide 2026 →