The Model Context Protocol (MCP), a widely used standard that lets AI models securely connect to external data sources and tools, is set to receive a significant technical update. The change, due next week, alters how the protocol manages session IDs, the tokens servers use to track ongoing conversations, making it easier for MCP servers to operate at large scale. Although invisible to end users, the update matters because it removes a major obstacle that has discouraged companies from building large-scale MCP integrations despite widespread interest in agentic AI.

Under the current system, a server hands a client a session ID that must be remembered by that specific machine, which creates problems for companies running services across many servers behind load balancers, since any server handling a request needs to recognise an ID issued elsewhere. The new version, whose spec has been public since May, moves to a "stateless" approach similar to how ordinary websites work, reducing maintenance burden and potentially cutting costs. Arcade's Nate Barbettini offered a detailed explanation of the change, which TechCrunch highlights as evidence that AI infrastructure development, unlike model training, still proceeds at the slower pace typical of standards-body consensus.