From Microsoft’s Copilot to Apple’s Apple Intelligence, both Windows and macOS are integrating AI features at an unprecedented pace. Even hardware manufacturers like AMD are branding their products as “AI CPUs.” AI and Large Language Models (LLMs) are being integrated into many products.

After learning about MCP yesterday, I predict that: Operating systems may soon offer system-level MCP services.

What does this mean?

In the future, the OS could be not just a platform for running applications, but a powerful dispatch center for AI agents. Whether it’s the built-in Copilot, Apple Intelligence, or third-party AI applications like llama.app, Cherry Studio, or Block Goose, they could all leverage this unified MCP service to gain direct, fine-grained control over the computer.

Of course, we have rudimentary versions of this today. Siri can create a calendar event for you, and Copilot can summarize a document. But this is not enough. These actions are more like pre-programmed shortcuts than truly autonomous agentic behavior.

In this future, you can tell an AI agent: “Organize all documents related to the ‘Sirius Project,’ create a password-protected archive, and email it to John Doe from the project team.” The agent, upon receiving the command, would securely access file permissions, execute the compression task, and invoke the mail client through the OS’s MCP service to complete the request. The process would be smooth and efficient.

Permissions and privacy are central to this vision. This power must be controlled. We can anticipate operating systems building a permission management system far more granular and stringent than the current “Allow access to Photos/Microphone” prompts. Users would be able to clearly see which agent requested what permission at what time, what operations were performed, and could grant or revoke these permissions at any moment.

Once operating systems open up this MCP capability, a new AI agent ecosystem will emerge. Will we see an official “AI Agent Store,” similar to the App Store, where users can safely download and purchase agents with various capabilities? Or will it evolve into a more open, decentralized model, allowing developers and enthusiasts to freely create and share their AI agents? Regardless of the model, competition and innovation will create a new era of AI agents.

The current AI features in our operating systems are just the beginning. The next step is transforming the OS itself into a “meta-platform” that provides the foundational support for countless AI agents. A system-level MCP service might be the key to this transformation.

The trajectory of technological development seems to be pointing in this direction.