MCP Gets Its Biggest Update Yet, and the Goal Is Letting AI Work at Company Scale
A key change to the open standard that helps AI tools talk to outside software removes a bottleneck that made large deployments tricky. Here is what changed and why it matters.
Key points
- The Model Context Protocol (MCP), an open standard for connecting AI systems to external tools, received its largest update since launch in 2024.
- The core change makes MCP "stateless", meaning requests no longer depend on a single dedicated server connection.
- The update was announced by lead maintainers David Soria Parra and Den Delimarsky, both of whom work at Anthropic, the AI company behind the Claude assistant.
- Removing the session dependency directly targets the scaling limits that previously made MCP difficult to deploy across large organisations.
Think of MCP, the Model Context Protocol, as a universal plug socket for AI. Just as a standard plug lets any appliance draw power from any wall socket, MCP is an open-source standard, meaning anyone can read, use or improve its rules, that lets AI systems connect to outside tools and data sources without each connection needing to be custom-built from scratch. A business might use it to let an AI assistant pull data from a calendar, a database, or a customer support system.
The standard launched in late 2024, and this week it received its most substantial update so far, as first reported by Ars Technica AI.
What actually changed?
The headline change is that MCP's core is now "stateless". Before this update, every conversation between an AI system and an external tool depended on a continuous, dedicated session tied to one specific server, a computer that handles requests. Lose that session and the connection breaks. Want to handle thousands of simultaneous users? You need to carefully route each one back to the same server instance, which is expensive and complicated.
Stateless design removes that dependency entirely. Each request carries everything it needs to be understood on its own, so any available server can handle it. This is the same principle that makes the ordinary web scale to billions of people.
For businesses, the practical effect is significant. Deploying AI tools across a company no longer requires building elaborate plumbing to keep sessions alive.
Who made this change?
The update was written by David Soria Parra and Den Delimarsky, the lead maintainers of the MCP specification. Both work at Anthropic, the company that originally created MCP before opening it to the wider community.
What does this mean for ordinary users?
Most people will never interact with MCP directly. It sits underneath the products built on top of it, so the benefit arrives quietly. AI assistants built on MCP-connected tools should become more reliable and faster, especially inside larger organisations where many people use the same system at once.
If your employer is piloting an AI assistant that connects to internal data, this update makes it considerably easier for the IT team to run that tool at full company scale without performance headaches.
Common questions
Do I need to do anything because of this update?
No. MCP is a developer-level standard. If you use a product built on it, any improvements flow through automatically when the developers update their software.
Is MCP only used by Anthropic products?
No. MCP is an open standard, so any company or developer can build with it. Anthropic created it and still employs the lead maintainers, but the specification is public and increasingly used across the AI industry.
