The ecosystem of artificial intelligence is undergoing a quiet but profound transformation. The Model Context Protocol (MCP), which began as a desktop solution for connecting AI applications to external systems, has evolved into an open standard. But its growth brought scalability issues. The July 28, 2026 revision (2026-07-28) introduces the most radical change: MCP becomes stateless, eliminating protocol-level sessions. According to Anthropic's maintainers, this decision responds to 'hard lessons' learned over the past two years. For companies looking to integrate AI agents into their workflows, this change opens new possibilities, but also requires rethinking existing architectures.
Originally, MCP used persistent sessions to maintain state between client and server. This worked well in local environments where a single developer interacted with coding tools. However, when moving MCP servers to the cloud to serve multiple enterprise clients, session management became a bottleneck. Each interaction required creating and supervising a session, complicating routing and load balancing. The solution has been to adopt a stateless model, similar to Anthropic's Claude Messages API. Now, information about protocol version, client identity, and capabilities travels in the _meta parameter of each request. The protocol no longer tracks sessions, greatly simplifying infrastructure.
This change is not only technical; it has significant business implications. Organizations that have adopted MCP to connect their AI systems with databases, APIs, or analytics tools can now scale more efficiently. By eliminating the need to maintain sessions, MCP servers can run behind standard load balancers, exactly like any other modern web service. This reduces operational costs and facilitates integration with cloud platforms such as AWS or Azure. In fact, many companies are already migrating their MCP deployments to the cloud to leverage the elasticity of these environments. Q2BSTUDIO's cloud services offer exactly that kind of support, helping companies design scalable and secure architectures for their AI agents.
But the transition to stateless is not without challenges. The underlying data transfer mechanism has been completely rebuilt. Although SDK updates should facilitate migration for those who have kept their libraries up to date, those who have built their own MCP implementations face a considerable reconfiguration effort. David Soria Parra, a technical staff member at Anthropic, has acknowledged this: 'If you built your own implementation, it's going to be a lot of uplift to make this correct.' Additionally, routing is now reflected in HTTP headers, allowing networking gear to route requests without inspecting the JSON-RPC body. This is positive for efficiency but requires adjustments in firewalls and proxies.
The revision also deprecates features that didn't see expected usage. Sampling, which allowed an MCP server to request text generation from the client's model, is removed due to its semantic complexity. The roots feature, which indicated relevant file system locations for a session, is also eliminated for being too niche. Logging is considered excessively chatty, and developers are advised to use stderr, stdio, or tools like OpenTelemetry. Instead, the protocol becomes more extensible: new features will be tested through official extensions with their own release schedule. For example, MCP Apps allows presenting interactive JavaScript applications, and Tasks (for long-running operations) has moved to an extension. This opens the door for each company to customize the protocol to their needs.
For Q2BSTUDIO, this evolution is an opportunity. As a company specialized in artificial intelligence and custom software development, we see how the stateless MCP simplifies the creation of AI agents that interact with enterprise systems. Our teams are helping clients adapt their MCP implementations, whether by updating SDKs, redesigning the network layer, or integrating the protocol with cloud solutions. Furthermore, the removal of sessions improves security by reducing the attack surface, as no server-side state storage is required. For those looking to protect their AI applications, our cybersecurity services offer audits and pentesting tailored to this new scenario.
The MCP ecosystem has grown impressively: over 97 million monthly SDK downloads and over 10,000 servers deployed. With the new stateless architecture, these numbers are expected to continue increasing, especially in enterprise environments requiring high availability. The ability to scale horizontally, combined with the protocol's extensibility, makes MCP a key piece for agent-based application development. Companies across all sectors, from finance to logistics, are exploring how to integrate AI assistants that securely and efficiently access their internal systems. In this context, having a technology partner who understands both the protocol and business needs is essential.
Beyond MCP, the move toward statelessness reflects a broader industry trend: the adoption of cloud-native principles in the AI layer. The same logic that made Kubernetes the standard for container orchestration is now applied to context protocols. Craig McLuckie, CEO of Stacklok and co-creator of Kubernetes, has noted that MCP's stateful nature was a byproduct of its local origin, but the future lies in stateless systems that allow platform teams to govern in a unified way. For companies using Power BI, for example, integrating AI agents through MCP can enable natural language queries on their dashboards, something that our BI services can facilitate.
In summary, the 2026-07-28 revision of MCP marks a before and after. The protocol abandons sessions to embrace a stateless model that simplifies deployment, improves scalability, and aligns with cloud best practices. Although migration can be complex for customized implementations, the long-term advantages are clear. At Q2BSTUDIO, we are prepared to accompany organizations in this transition, offering custom application development, cloud integration, cybersecurity, and AI solutions that leverage the full potential of the new MCP. The future of AI agents is stateless, and it is already here.
