Hajra Naz
The Model Context Protocol (MCP) is one of the most important standards in AI. It helps AI models connect to external apps and data safely. This allows chatbots to access calendars, databases, and business tools. Without MCP, developers would need to build a separate connection for every service.
A major update to the protocol is arriving next week. Most users will never notice the change. However, it could have a big impact on developers and companies building AI applications.
The official specification for the update has been available since May. On Monday, startup Arcade explained the changes in simple terms. The company has spent the last two years building tools that help AI agents work inside businesses. Its platform lets AI connect securely with services such as Gmail, Slack, and Salesforce.
Arcade believes the biggest challenge isn’t today’s AI models. Instead, it says the supporting infrastructure is still catching up. The company raised $60 million in June based on that idea. It argues that many AI agents fail because the systems around them are not ready. The latest MCP update is designed to solve part of that problem.
The biggest change involves session IDs. These are small pieces of information that help a server remember an ongoing conversation. They tell the server that a new request belongs to the same user instead of starting a new session.
Today, the process works like this. An AI client, such as Claude, connects to an MCP server for the first time. It sends a greeting that includes its version and capabilities. The server replies with its own details. It also creates a session ID. After that, the client includes the same session ID with every request.
Read More: Anthropic Unveils Claude Sonnet 5, a Lower-Cost AI Model for Agentic Tasks
Sometimes the session ID expires. When that happens, the client must detect the problem. It then asks the server for a new session ID before continuing the conversation.
This system works well on a single server. It becomes much harder at a large scale.
Imagine a company serving millions of users every day. Their traffic is spread across many servers. A load balancer sends each request to whichever server is available. That server may even be in a different region.
The problem is simple. One server creates the session ID. Another server may receive the next request. Every server must recognize the same session ID. Keeping them synchronized adds extra complexity. It also runs counter to how load balancers are designed to operate.
In other words, the current system assumes one server remembers each user. Large companies don’t work that way. They distribute traffic across dozens or even hundreds of servers. Those servers do not automatically share session information. As a result, developers must build extra systems just to keep everything running smoothly.
That has become a major challenge for companies using MCP at scale. It is also one reason why more businesses haven’t launched large first-party MCP integrations, despite growing interest in AI agents.
The new version changes that approach. Instead of depending heavily on server-side session tracking, MCP will move toward a more stateless system. This is similar to how many modern websites already operate.
The new design should make servers easier to manage. It should also reduce maintenance work. Running large AI systems could also become cheaper.
Most users will never notice these improvements. Their AI assistant will continue working as usual. Behind the scenes, however, developers should find it much easier to build reliable AI services that can support millions of users.
The update also highlights an important point about AI development. New models continue to improve at an incredible pace. The infrastructure behind them moves much more slowly. Industry standards take time to develop. They require testing, discussion, and agreement before they become official.
That slower process may not grab headlines. Even so, it plays a critical role in the future of AI. Strong standards like MCP make it easier for developers to build secure, reliable, and scalable AI applications.
