The Model Context Protocol is receiving a major update next week to simplify how AI agents connect to external services like calendars and databases. This change improves the underlying infrastructure for AI interoperability without necessarily altering the experience for end users.
The technical fix
Arcade, a two-year-old startup founded by Nate Barbettini, released a clear explanation of the changes on Monday. The company raised $60 million in June based on the belief that AI agents often fail not because the models are weak, but because the surrounding infrastructure is not ready.
Current versions of the protocol rely on session IDs to maintain context. When a client like Claude connects to a server, it sends a greeting containing its version and capabilities. The server replies with its own details and returns a session ID. The client then sends this ID with every subsequent request so the server knows the conversation is continuous.
Barbettini noted that if the ID expires, the client must request a new one to continue. This process creates problems when deployed at scale.
Picture a real deployment. You’re running a server for millions of users, behind a load balancer whose entire job is to route each request to whatever server in the farm is free, sometimes in a different region. Now every one of those machines has to know about a session ID that some other machine handed out. It’s not impossible, but it’s a serious pain, and it fights the load balancer instead of working with it.
The current setup assumes a single server remembers the user. In reality, companies spread traffic across dozens of servers that do not communicate by default. This forces MCP servers to perform extra work to track identities, creating a headache for anyone trying to run large-scale integrations.
The new system adopts a looser, stateless approach to session IDs on the server side. This mirrors how most ordinary websites operate. The change should make the system easier to maintain and cheaper to run at scale.
What it means
This update highlights that not all AI development moves at breakneck speed. While model training races ahead, the technical infrastructure required by those models often follows the slow pace of standards-body consensus. The ecosystem is still improving, albeit gradually.
