Latest MCP Specification Eliminates Session Management

The latest specification of the Model Context Protocol (MCP), which is becoming established as a standard specification for AI agent and tool integration, has removed the session management mechanism from the protocol itself. AWS has published technical details of this specification change, explaining that the requirement for sticky sessions (fixed routing) that was previously necessary when operating remote MCP servers is no longer needed.

MCP is a common communication standard that allows AI models to exchange information with external tools and data sources. In the previous MCP specification, a mechanism for managing a series of user operations as a "session" was built into the protocol itself. The server side needed to maintain session state, and a "sticky session (fixed routing)" configuration was required to ensure that requests from the same user always reached the same server. This became a constraint in system design, making flexible operations in cloud environments difficult.

In the latest specification, this design has changed, and MCP servers now operate in a "stateless" configuration that does not maintain session state. Each request is processed independently, so it can function properly regardless of which server instance it is routed to. Issues such as session state storage, management, retry, monitoring (observability), and idempotency (the property that repeating the same operation does not change the result) are now handled by the application layer or underlying infrastructure outside the protocol.

As for how this change actually affects things, the most direct benefit is that "horizontal scaling (horizontal expansion)" becomes easier. Previously, due to sticky session constraints, simply adding more servers could not always distribute traffic evenly. With statelessness, each request can be distributed to any server, allowing load balancing and auto-scaling in cloud environments like AWS to function naturally.

As background, MCP is a specification proposed by Anthropic and is increasingly being adopted across the industry as a "common language" for AI agents to work with multiple tools and external APIs. Many companies have begun deploying services using MCP on the cloud, and scalability and operational cost reduction have become important practical issues. This specification change is positioned as a response to such practical needs.

On the other hand, the shift of session management responsibility from the protocol to the application layer increases implementation flexibility, but also means developers need to design state management and retry mechanisms themselves. As a tradeoff between operational simplification and design responsibility, consideration of this point is necessary in actual system implementation.

AWS's publication of technical explanations of this specification change suggests that the company positions MCP as an important foundational technology for its cloud services. The evolution of MCP specifications and how various cloud companies respond to it have the potential to influence the design approach of AI agent infrastructure, and future specific implementation cases and moves by other cloud vendors are worth noting.