The Model Context Protocol (MCP) has received a massive update, transitioning from a stateful model to a stateless architecture. This fundamental technical shift transforms the protocol from a local-use tool into a full-fledged infrastructure ready for high-load cloud and edge environments.

image
image

What Happened

According to the new specification dated July 28, 2026, MCP has stopped relying on session state preservation. Requests are now independent, allowing the use of standard HTTP load balancers for horizontal scaling. The update also introduces mechanisms for asynchronous tasks (Tasks), the implementation of MCP Apps within isolated iframes, and a transition to using Client ID Metadata Documents (CIMD) to enhance security.

Context

Previously, MCP functioned primarily as a protocol for managing local sessions, which imposed significant limitations on its use in distributed systems. The transition to a stateless architecture eliminates the need for complex state management on the server side, allowing for the use of familiar cloud deployment patterns.

Why It Matters for the Industry

For the industry, this means the transformation of MCP into a scalable, industrial-grade infrastructure suitable for cloud-native and distributed architectures. The protocol gains the potential to become the de facto standard for interaction between LLMs and external data and tools in the enterprise segment, similar to the role of HTTP in web technologies.

Why It Matters for Users

Developers will be able to deploy tools more easily in serverless and edge environments without the need to maintain complex sessions. Users will benefit from faster and more reliable cloud-based AI tools, while the emergence of MCP Apps in isolated environments will provide a higher level of security when interacting with external interfaces.

What Is Not Yet Known / Limitations

The transition to a stateless architecture may require significant logic changes on the side of existing clients and servers that were previously dependent on session state management.

Sources

Author

Look at AI, Editorial Staff