Ok so this one sounds techy, but stay with me, because it quietly affects the AI apps more and more people use every day. There is a hidden piece of technology called the Model Context Protocol. It is what lets an AI helper safely reach into your other apps, like your calendar or your files, and actually do things for you. That hidden piece is getting its biggest change since it was created. The new version becomes final on July 28, 2026. A tech news site called TechCrunch explained it on July 20, 2026.
People usually just call it MCP for short. Let me explain it like you have never heard of it, because most people have not.
What MCP actually is
Imagine you have a smart helper on your phone. You want it to book a meeting, check your files, or pull up your bank record. For the helper to do that, it needs a safe way to reach into each of those apps.
MCP is that safe way. Think of it as a shared power socket. Before a shared socket existed, every app needed its own special plug, made by hand, one by one. That was slow and messy. MCP is like agreeing on one standard socket shape, so the helper can plug into many apps the same way.
Two quick words, explained. When I say MCP is an “open standard,” I just mean it is a free shared rulebook that anyone is allowed to use. And the company that created it is Anthropic, the maker of the AI helper called Claude.
So MCP is basically the wiring between an AI helper and the rest of your digital life. When you hear about AI that can do tasks for you, not just chat, this wiring is a big part of what makes that possible.
What is changing on July 28
Now the update. To explain it, let me show you how MCP works today, using a simple example.
Picture walking into a bank. The first time, you show your ID. The clerk gives you a numbered ticket and says, “Keep this, it proves who you are.” After that, every time you speak to that clerk, you show the ticket, and they remember you.
MCP works the same way right now. The AI helper connects, proves who it is, and gets a little “ticket” so the service remembers the conversation. In tech words that ticket is called a session ID, but ticket is all you need to picture.
That works fine with one clerk. The trouble starts when the place gets huge.
Now imagine a giant bank with a hundred clerks, and every visit you might get sent to a different one. Your ticket came from clerk number 4, but now you are standing in front of clerk number 71, who has never seen you. For that to work, all hundred clerks have to constantly share notes about every ticket they handed out. That is a lot of extra effort, and it slows everything down.
Big AI services have exactly this problem. To serve millions of people, they spread the work across many machines, and those machines do not naturally share notes.
So here is the fix. The new MCP drops the ticket system on the service side. Instead, each request quietly carries its own proof of who it is, every time. So it does not matter which “clerk” you land in front of. This tidier approach is the “stateless” design you may see mentioned, and stateless just means no ticket to keep track of. TechCrunch says this should make these services much easier to run, and cheaper too, when serving huge numbers of people.
Why this was holding things back
Here is a thing I didn’t fully realise. That ticket headache is one reason big companies have been slow to build their own large AI connections this year, even with all the AI hype. The wiring was fighting them whenever they tried to grow.
A company called Arcade explained the old problem well, and TechCrunch quoted them. Their engineer said the old way “fights the load balancer instead of working with it.” A load balancer is just the traffic guide that sends each visitor to whichever machine is free. So the old system was working against the very helper meant to handle big crowds.
A couple of honest notes
Let me keep this fair and clear.
First, you will not see this as a new button in your app. It is a behind-the-scenes change. The report says everyday users likely will not notice it directly. The benefit is hidden, in how smoothly and cheaply these AI connections run.
Second, it is not fully final yet. A near-final version is out now, but the official final version lands on July 28, 2026, and the plan has been public since May 2026 so people can give feedback. So it is close, but not done.
Third, bigger and more open wiring can bring new safety questions. Separate reports this suggests the change fixes some old weak spots but also gives builders new things to protect. I am only passing on what those sources say, so treat the safety part as an open topic, not a final answer.
So see this as an important groundwork fix, quietly setting up what comes next.
Why this could matter
If we analyze, we can see the choice the MCP team made. They gave up the neat “ticket that remembers you” in exchange for a system that runs smoothly across many machines at once. That choice reportedly buys real gains. It makes big AI setups easier to run, cheaper to grow, and in some ways simpler to keep safe, which could finally let more companies build AI helpers that truly plug into your apps. But it has a cost. Taking the memory out of the service means the builders now have to carry that “who is this” proof themselves, and a bigger, more open system means more safety details to watch. In plain words, they traded cozy but fragile for plain but able to grow, and that trade is the whole reason this quiet update could speed up the AI tools you actually use.
So here is the real question for you. Would you want an AI helper that can plug into all your apps and get real tasks done, even if it means trusting more hidden wiring behind the scenes? And how closely should that wiring be watched for safety as it grows?
If you like tech news explained in plain simple words like this, subscribe to the VulpisLab newsletter for a small daily dose.
Quick questions people are asking
What is the Model Context Protocol?
It is a free shared system, created by Anthropic, that lets AI helpers safely connect to your other apps and tools, like your calendar, files, or a company’s systems.
What does MCP stand for?
MCP stands for Model Context Protocol.
What is changing in the July 2026 update?
MCP is dropping its “ticket” memory system on the service side. This makes big AI services easier and cheaper to run for millions of people.
When does the update go live?
The final version lands on July 28, 2026. A near-final version is already out now.
Will regular users notice the change?
Probably not directly. It is a behind-the-scenes fix, so the benefit shows up in how smoothly AI connections run, not as a new feature you tap.
Who reported this?
TechCrunch explained it on July 20, 2026, using an explainer from a company called Arcade, along with the official MCP plan.
Source: TechCrunch, “AI’s most important protocol is getting a little bit easier to use,” July 20, 2026: https://techcrunch.com/2026/07/20/ais-most-important-protocol-is-getting-a-little-bit-easier-to-use/ . Official plan: Model Context Protocol blog, release candidate for 2026-07-28: https://blog.modelcontextprotocol.io/posts/2026-07-28-release-candidate/