MCP Drops Sessions in July 28 Release Candidate as Stateless Design Targets Cloud Scale
Updated
Updated · InfoWorld · Jul 24
MCP Drops Sessions in July 28 Release Candidate as Stateless Design Targets Cloud Scale
3 articles · Updated · InfoWorld · Jul 24
Summary
July 28’s MCP release candidate removes protocol-level sessions, shifting the AI tool-connectivity standard to a stateless architecture aimed at production cloud deployments.
Under the new design, each request carries its own context, letting any server handle it without sticky routing and making MCP easier to scale across standard cloud infrastructure.
New features include Multi Round-Trip Requests, routable transport headers, OAuth 2.1 and OpenID Connect authorization, interactive MCP Apps, and deterministic caching to cut latency and token costs.
A 1-year transition period keeps deprecated features working, but dropping Sampling could force teams to redesign trust boundaries, model-provider connections, authentication, and billing flows.
Updated Python, TypeScript, Go, and C# SDKs support both old and new versions, easing migration even as enterprises audit hidden session dependencies in gateways, scripts, and monitoring.
Will the deprecation of protocol-level sessions in the upcoming MCP update secretly break your enterprise AI gateways before the migration window closes?
As MCP removes client-side LLM sampling, how will enterprise teams handle the sudden shift of billing and authorization boundaries to the server?
With MCP going stateless on July 28, will the payload overhead of explicit context negate the scaling benefits for complex AI agents?