Going into this week, we knew we’d be getting a ton of Gemini AI news from Google’s I/O 2025 event, and that’s exactly what happened. Google made several key announcements for its AI products, and the future of Gemini seems brighter than ever.
But we got even more than that, as Microsoft hosted its annual Build conference before I/O, where it announced its own Copilot AI innovations for Windows.
Then there was OpenAI’s surprise announcement about Jony Ive’s io startup, which will manufacture the first of many ChatGPT products in the coming years. On top of that, OpenAI announced several improvements for its AI services.
All these separate product events had one thing in common. All three companies talked about something called MCP, short for Model Context Protocol, which is coming to Copilot, Gemini, and ChatGPT.
The name itself doesn’t mean much to most people, and we’ll probably forget about MCP after this week’s flood of AI announcements. But MCP, described as the USB-C port of AI apps, represents a major step forward. It will allow chatbots to perform more advanced assistant tasks, whether it’s accessing personal data or controlling apps and devices.
The comparison to USB-C ports is meant to highlight how useful MCP is for the industry. MCP is a standard communication protocol between AI chatbots, apps, operating systems, databases, and anything else you might want to get data from or control with AI. The wide range of AI agents available across platforms will also rely on MCP tech to get things done for you.
But unlike USB-C, which you see and use daily, MCP isn’t something you’ll think about. You won’t ask if it’s built into your AI model, or whether your devices or favorite apps support it. It’ll just be there, working behind the scenes so your AI of choice can help you better.
For example, you might tell the AI to message your mom with your estimated arrival time. The AI will access your messaging platform via an MCP server and handle the task for you. You won’t even realize that this protocol is powering the communication between the chatbot and the program.
OpenAI shows a few examples of MCP integration in popular apps that let ChatGPT interact with them on your behalf. One example shows how you can add items to a Shopify cart by telling ChatGPT to do it.
You could also ask the AI to pull information from your calendar or email. In this area, Google’s Gemini has a big edge. Google can tap into all its apps to let Gemini surface exactly what you need. Apple’s Siri could do the same, but that Apple Intelligence feature has been delayed until next year.
With MCP tech, though, other AI chatbots might gain access to the third-party apps you use on your phone or computer.
MCP will also let AI products control your computer. For instance, you could tell Copilot to delete all files in your downloads folder. MCP will enable the AI to find the folder, view its contents, and execute the delete command. And it doesn’t have to be the built-in AI doing the work. Microsoft showed The Verge how an AI like Perplexity can search through files on a Windows 11 machine for a specific topic. MCP tech is what makes this kind of assistant behavior possible.
When I looked at Google’s big Gemini announcements this week, I highlighted Google’s advantage over OpenAI right now, as Gemini can offer features that ChatGPT can’t. Google’s broad ecosystem of apps helps Gemini become a more personal assistant, since it can connect its AI and agents to those apps safely.
But with MCP in play, AIs like ChatGPT might be able to connect to Gmail and Google Maps on their own, as long as you allow it. It won’t be as seamless as Gemini’s integration with Gmail, but it should still work well and let ChatGPT offer similar help. Google also supports MCP as of Tuesday.
The MCP standard is designed to protect user privacy, with AI only accessing the data it’s permitted to. You’ll likely see pop-ups and prompts each time you allow an AI app to access data from another app.
The protocol should also help keep data safe, though it does introduce new potential threats. Microsoft has an entire blog post on MCP security. Since MCP is still in early development, there are security risks. That’s why Microsoft is releasing MCP support in preview mode for developers only.
MCP probably won’t come up often in regular conversations unless you’re a developer. But we’ll see its effects over time, as AI evolves into the kind of personal assistant that can access data from any app you use and control both apps and your operating system.
Put simply, MCP tech will be critical for OpenAI’s first io ChatGPT device to function. The product is expected to be a pocketable, non-smartphone gadget that’s aware of your context and ready to help whenever you need it.
