The Agentic AI Foundation, a Linux Foundation project, has released a major update to the Model Context Protocol (MCP), the open standard that lets AI agents connect to external data sources and tools. The revision, described by co-creator David Soria Parra as MCP's most significant change since remote MCP launched, overhauls the protocol's architecture to make it far easier for enterprises to run at scale, addressing complaints about the complexity of deploying it in production environments.

The new version, dated 2026-07-28, removes the legacy stateful architecture so MCP servers can run behind standard load balancers on ordinary Kubernetes and DevOps tooling, rather than requiring sticky routing or shared state. It also introduces a formal Specification Feature Lifecycle and Deprecation Policy guaranteeing a minimum 12-month gap between feature deprecation and removal, a security fix (SEP 2468) requiring validation of an issuer parameter to prevent OAuth Mixup Attacks, support for centralised identity management via an Enterprise Managed Authorization extension, and a shift of long-running "tasks" to an asynchronous, crash-resilient model.