Model Context Protocol (MCP) is one of the fundamental building blocks of AI interoperability, providing AI models with a secure way to access external data sources and services. Rather than having engineers build custom pipes for every connection, this is plumbing that allows your chatbot to access calendars, databases, or internal tools. Next week, the protocol will receive a significant update, which may not be noticeable to end users, but could lead to major changes in how the ecosystem develops.
Official specifications for the new version have been available since May, but we received an unusually clear explanation of the changes from Arcade staff on Monday morning. Essentially, MCP makes it easier for servers to operate at larger scales by changing the way they handle session IDs, which are little tokens that servers use to remember, “Oh, this is the same conversation from five seconds ago.”
Arcade’s Nate Barbettini said:
(On current systems) When an MCP client like Claude first connects to a server, it sends “hello”. My name is Claude. Here is my version. Here is my function. The server responds with its own function and returns a session ID. From then on, the client sends that session ID for every request so the server knows it’s the same conversation. Sometimes the ID expires and the client needs to check it, request a new ID and move on.
Imagine a real deployment. You’re running servers for millions of users, and the entire operation is behind a load balancer that routes each request to free servers in the farm, and sometimes to servers in other regions. Now every machine needs to know about the session ID passed by another machine. It’s not impossible, but it’s a significant pain and fighting the load balancer instead of using it.
This means that while the current setup assumes one server remembers you, real-world companies essentially distribute their traffic across dozens of servers that don’t talk to each other, so today’s MCP servers have to do extra work to keep track of who’s who. This has been a major headache for anyone running MCP servers at scale, and is one of the reasons why, despite all the hype about agent AI this year, more companies haven’t been able to launch their own MCP integrations at scale.
In the new system, the protocol will take a looser, “stateless” approach to server-side session IDs, similar to how most ordinary websites already work. This makes the overall system much easier to maintain and theoretically cheaper to run at scale.
This is all pretty technical, but it’s an important reminder that not all aspects of AI development are moving at breakneck speed. While model training is ongoing, much of the technical infrastructure required for that model is still subject to the slow log rolling of standards body consensus. That’s actually happening. It’s just a little slower!
If you purchase through links in our articles, we may receive a small commission. This does not affect our editorial independence.
