The Model Context Protocol (MCP) serves as a foundational element for AI interoperability, allowing AI models to securely connect with external data sources and services. Think of it as the underlying infrastructure that enables a chatbot to access your calendar, database, or internal tools—without requiring developers to build custom connections for each integration. Next week, this protocol is receiving a significant update. While end users may not notice the change directly, it could have a major impact on how the ecosystem evolves.

The official specification for the new version has been publicly available since May, but on Monday morning, the team at Arcade provided an unusually clear explanation of what’s changing. At its core, MCP is altering how it handles session IDs—the small tokens servers use to recognize that a request belongs to the same ongoing conversation. This adjustment allows servers to operate more efficiently at larger scales.

As Arcade’s Nate Barbettini explains:

[Under the current system] The first time an MCP client like Claude connects to a server, it sends a “hello” message: I’m Claude, here’s my version, here are my capabilities. The server replies with its own capabilities and provides a session ID. From that point on, the client includes that session ID with every request so the server knows it’s part of the same conversation. Sometimes the ID expires, forcing the client to notice, request a new one, and continue. Now, imagine a real-world deployment. You’re running a server for millions of users behind a load balancer whose sole job is to route each request to whichever server in the farm is available—sometimes in a different region. Every one of those machines must be aware of a session ID that another machine handed out. It’s not impossible, but it’s a significant hassle, and it works against the load balancer rather than with it.

In other words, the current setup assumes a single server remembers each user, but real companies distribute traffic across dozens of servers that don’t naturally communicate. This forces today’s MCP servers to do extra work just to keep track of who’s who. That’s been a major headache for anyone running an MCP server at scale, and it’s part of the reason we haven’t seen more companies roll out large-scale, first-party MCP integrations, despite all the hype around agentic AI this year.

Under the new system, the protocol will adopt a looser, “stateless” approach to session IDs on the server side—similar to how most ordinary websites already function. This shift should make the entire system much easier to maintain and, theoretically, cheaper to run at scale. While all of this is highly technical, it’s an important reminder that not every aspect of AI development moves at breakneck speed. As model training races forward, much of the technical infrastructure those models depend on still progresses at the slow, deliberate pace of standards-body consensus. It’s happening; it’s just a bit slower.