The Model Context Protocol (MCP), the plumbing that lets chatbots reach into calendars, databases, and internal tools without custom pipes, is getting a quiet but significant overhaul. Next week, a new version of the protocol will go live that abandons the current session ID system in favor of a stateless approach, a change that could unlock large-scale deployments of AI agents inside real companies.
MCP gives AI models a secure way to access external data sources and services. The official spec for the new version has been public since May, but the implications are only now being fully explained by companies building on the protocol. Arcade, a two-year-old startup that raised $60 million in June, provided a detailed explanation of the changes on Monday morning.
Arcade’s entire business is built around getting AI agents to function inside real companies, securely connecting them to tools like Gmail, Slack, and Salesforce. The company’s thesis is that most AI agents fail because the infrastructure is not ready, not because the models are weak. The current MCP system, they argue, is a prime example of that infrastructure problem.
Under the current system, a client—say, an AI model like Claude—sends a "hello" message with its version and capabilities. The server replies with its own capabilities and hands out a session ID. The client then sends that session ID on every subsequent request so the server knows it is the same conversation. The session ID can expire, forcing the client to request a new one.
This setup works fine for a single server handling a handful of conversations. But in a real deployment, a server for millions of users sits behind a load balancer that routes each request to whatever server in the farm is free, sometimes in a different region. Each machine must know about the session ID handed out by another machine. The current setup assumes one server remembers the user, but real companies spread traffic across dozens of servers that do not talk by default.
Nate Barbettini, founding engineer at Arcade, described the current system’s core assumption. "Ah, this is the same conversation as five seconds ago," he said, mimicking the server’s internal logic. That assumption, he explained, forces MCP servers to do extra work to track who is who. The current session ID setup fights the load balancer instead of working with it.
This has been a headache for anyone running an MCP server at scale. It is part of the reason few companies have shipped large-scale first-party MCP integrations, despite the agentic AI hype this year.
A Stateless Fix
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
The new system takes a looser, "stateless" approach to session IDs on the server side. Instead of requiring the server to remember which session belongs to which user, the protocol will treat each request more independently. This is similar to how most ordinary websites work, where the server does not need to hold a persistent memory of every visitor.
The change should make the whole system easier to maintain and cheaper to run at scale. For end users, the update might not be noticeable, but it could make a big difference in the ecosystem’s development. It removes a key barrier that has kept companies from building and deploying MCP-powered agents to large user bases.
Infrastructure Moves Slowly
The update underscores a broader reality in AI development. Not every part of the field moves at breakneck speeds. While model training races ahead, with companies spending $330 million or $100 million on a single training run, the infrastructure that connects those models to real-world data is subject to slow standards-body consensus.
"It really is happening; it's just a little slower," Barbettini said, referring to the pace of infrastructure standardization. The MCP protocol is a basic building block of AI interoperability, and changes to it require careful consideration from the community.
Arcade, which has been operating for two years and raised $60 million in June, is betting that solving these infrastructure problems is the key to making AI agents actually useful. The company connects AI agents to tools like Gmail, Slack, and Salesforce, and its thesis is that the infrastructure needs to be ready before the agents can work.
The stateless session ID update is a small but critical step in that direction. By removing the burden of session tracking from the server, the protocol becomes more scalable and more aligned with how modern web infrastructure already works. It is a fix that addresses a real pain point for developers, and it could accelerate the deployment of AI agents in enterprise settings.