The Model Context Protocol (MCP) — started by Anthropic and now under the Linux Foundation — just got its biggest update yet. The universal protocol that lets AI agents connect to software, data, and tools has been completely rearchitected as stateless.
Before, an agent that connected to one MCP server had to keep talking to that same machine. If the server restarted or traffic got routed elsewhere, tasks could break.
The new version ditches the session mechanism. Every request now carries all the info it needs, so any server can pick it up. Enterprises can use standard load balancers and cloud services, making large-scale agent deployment much simpler.
MCP also beefed up authentication and made interactive interfaces and long-running tasks official extensions. Agents can now display forms and charts directly, and can keep working through long tasks even if they get disconnected.