Major Update to AI’s Key Protocol Simplifies Server Scaling

The Model Context Protocol (MCP) is a fundamental building block for AI interoperability, providing a secure way for AI models to access external data sources and services. It acts as the plumbing that allows a chatbot to reach into calendars, databases, or internal tools without requiring custom integrations. Next week, the protocol will receive a major update that, while perhaps unnoticeable to end users, could significantly impact the ecosystem's development.

The new specification has been publicly available since May, but a clear explanation of the changes came on Monday from Arcade, a two-year-old startup that has built its business around making AI agents work in real companies by securely connecting them to tools like Gmail, Slack, and Salesforce. Arcade raised $60 million in June, operating on the premise that most AI agents fail not because of weak models but due to inadequate infrastructure—a problem this update aims to address.

Essentially, MCP is changing how it manages session IDs, the tokens servers use to recognize ongoing conversations. In the current system, when a client like Claude first connects to a server, it sends a greeting with its version and capabilities. The server responds with its own capabilities and returns a session ID. The client then includes this ID in every subsequent request. If the ID expires, the client must obtain a new one. In real-world deployments, servers handling millions of users sit behind load balancers that distribute requests across multiple servers, sometimes in different regions. The current setup forces every server to track session IDs issued by others, which is cumbersome and works against the load balancer's purpose.

The update introduces a stateless approach to session IDs on the server side, similar to how most websites operate. This will make the system easier to maintain and, theoretically, cheaper to run at scale. While technical, the update highlights that not all AI development moves at breakneck speed—model training races ahead, but the necessary infrastructure still progresses through slower standards-body consensus.