Model Context Protocol moves to stateless session IDs to ease server scaling

The Model Context Protocol (MCP), a foundational protocol for AI interoperability, is preparing a significant update to how it manages session IDs. The current system requires servers to track session state across distributed systems, which complicates scaling for large deployments. Under the new approach, MCP will adopt a stateless model for session IDs on the server side, similar to how most websites handle user sessions today.

This change is intended to reduce the operational overhead for companies running MCP servers at scale. Previously, servers had to maintain session continuity across load balancers and distributed infrastructure, often requiring custom coordination to ensure requests from the same client were consistently routed to the same server. The stateless approach removes this requirement, simplifying deployment and potentially lowering costs for organizations integrating MCP into their workflows.

Arcade, a two‑year‑old startup focused on enabling AI agents to function within enterprise environments, emphasized the practical impact of the update. Arcade’s business is built around helping companies securely connect AI agents to tools like Gmail, Slack, and Salesforce. The company noted that many AI agent deployments struggle not due to model limitations but because the surrounding infrastructure remains underdeveloped.

The updated MCP specification has been public since May, but the explanation of its implications for server‑side scaling was provided by Arcade. The change reflects a broader trend in AI infrastructure: while model capabilities advance rapidly, foundational protocols and standards evolve more gradually, often through consensus‑driven processes.