Control Your PC from Claude: An Intro to MCP Automation

Automation used to mean "a person sets up the steps in advance, then runs them." You recorded macros, wrote scripts, pressed play — the human was always the lead, and AI at most helped write text. But recently a mechanism has appeared that greatly changes this relationship: MCP (Model Context Protocol). With MCP, an AI assistant like Claude can directly call and operate the tools on your PC.

In other words, a world is becoming real where you just say, in plain language in a chat, "summarise yesterday's sales and make the usual report," and the AI calls your automation tool and actually runs the macro. This article explains, without requiring specialist knowledge, what MCP is, why it fundamentally changes how automation works, and how to connect it and try it on your own PC.

What is MCP?

MCP (Model Context Protocol) is a common convention (protocol) by which AI assistants call external tools and apps. Until now, AI could basically only read text and return text. MCP breaks that wall, letting the AI tell a real tool "perform this operation" and receive the result.

When an app exposes an "MCP server," the AI can discover the list of operations that app offers. For an automation tool, operations like "run a macro," "run a script" and "start recording" become visible to the AI. From there, the AI simply picks and calls the right operation for the user's intent. A single "do X" in chat is translated into a real click or process on screen — that is the essence of MCP.

Why MCP changes automation

Traditional macros had one weakness: a person had to remember which macro does what. The more macros you have, the more tedious it becomes to find and run the right one. MCP removes this. You just describe your goal in plain language, and the AI picks the macro or script that matches your intent.

Even bigger is that the AI can generate and run a new script on the spot. Asked for something no existing macro covers, the AI can assemble an SCR script from a natural-language request, run it, read the result, and fix it itself if it doesn't work. This is a wholly new form of automation, a big step beyond a static list of recorded macros — an evolution from "make a fixed task faster" to "say what you want each time and it happens."

What you can do with MCP

With MCP integration, things like the following become possible through natural conversation alone:

• • Ask "make my usual morning report" and have a macro run
• • Instruct "rename all the images in this folder" in plain language
• • Have the AI script and run a task no existing macro covers, on the spot
• • Let the AI check the result and, if there's a problem, fix and re-run it

All of this happens as a natural extension of talking with the AI. No need to open a dedicated console to hunt for a macro, or to recall complex settings each time. The experience of "ask and it happens" greatly lowers the barrier to automation.

Actually getting connected

To try MCP integration you need an MCP-capable AI client (such as Claude Desktop or Cursor). And with an automation tool that has an MCP server, you're set. makuroku ships an MCP server built in from the start, exposing 18 tools to the AI that cover recording, playback, scripting and more.

Connecting is very simple. Launch makuroku and its MCP server starts automatically. Copy the connection URL from the "MCP" tab in settings and paste it into your AI client's config file — once you're used to it, it takes under a minute. And this feature is available up to 20 times a month even on the free plan, so you can experience AI-driven automation risk-free before committing.

Concrete scenarios for MCP integration

MCP integration shines when you want to ask for "work that differs slightly each time." For fully fixed tasks, running a traditional macro by shortcut is enough. But requests whose conditions change each time — "today, rename only the .png files in this folder" — are hard to prepare a macro for in advance. Here the value of the AI understanding the request and assembling a script on the spot comes alive.

Work needing judgement across multiple steps also suits dialogue with an AI. You can convey, in plain language, requests with situational branching like "if an error appears, check it, fix it if you can handle it, and report to me if you can't." The AI reads the result, fixes it itself if needed, and asks the human when unsure — this "automation that proceeds in cooperation with a human" is a wholly new way of working that traditional macros lacked.

Security and cautions when delegating to AI

That an AI can operate a PC directly is convenient but also calls for care. First, it matters to know which app's MCP server you're connected to and to use only trustworthy tools. The scope of operations exposed to the AI is limited to what that tool provides, so understanding "what it can and cannot do" brings peace of mind.

Second, for irreversible operations — permanently deleting files, sending data externally, or payments — it's safest to have a human confirm before execution rather than leave it to the AI. In many cases you can design a confirmation step before important operations. And for work handling confidential data, check where that data is stored and sent. Choosing a tool designed so recorded data stays local on your device and isn't auto-sent to the cloud, like makuroku, shrinks this worry. Harness the AI's convenience while a human holds the key points — this balance is the key to using AI integration with confidence.

Choosing between traditional automation and AI integration

AI integration is attractive, but leaving all work to the AI isn't always optimal. Traditional macros and scripts, and AI integration, each have situations they're good at. For routine work done by the exact same steps every day, running a pre-built macro by shortcut or schedule is faster, more reliable and cheaper. You needn't route fixed work through an AI on purpose.

AI integration, by contrast, shines for "work whose content changes slightly each time" and "work involving multiple judgements." The flexibility of the AI assembling a process to fit the moment just from a plain-language request comes alive precisely for work where preparing every pattern in advance is hard. So the two are not rivals but complementary. Solidify the routine parts with reliable macros, and leave the changing parts and those needing judgement to the AI — this division is the most efficient in practice.

The key is not to over-trust the AI as a "magic box." The AI shows its power precisely because there is a foundation of reliable automation — recorded macros and verified scripts. With a solid foundation the AI can conduct by cleverly combining it, but without one, what the AI can do alone is limited. That is exactly why the steady approach is to first master basic recording and scripting, then layer AI integration on top.

What you need to start with MCP integration

To actually start with MCP integration you need two main elements. One is an MCP-capable AI client — an app that can call external tools via MCP, such as Claude Desktop or Cursor. The other is a tool with an MCP server — the app operated by the AI. Only when these two are in place does the "AI operates the PC" integration come together.

The connection procedure itself is easier than expected. Launch the automation tool and write the connection info shown there (a URL and token) into the AI client's config file. Once set up, you can call the tool naturally within your conversation with the AI. It may sound hard, but following the official guide, even a first-timer can usually connect in a few minutes.

A new way of working that AI integration opens up

Once AI can operate a PC directly, how you engage with automation itself changes. Until now, a person thought about "what to automate," built the steps and ran them. But with AI integration, you just convey "what you want to achieve," and the AI works out how to achieve it. This flip of roles greatly lowers the barrier to automation.

For instance, "work not worth automating but quietly tedious" tended to be left manual, because building a macro felt like more effort. But with AI integration, you just ask "handle this" on the spot. With almost no setup effort, even the "small tasks" that were outside automation's scope can be handed over casually. Automation shifts from being for a few enthusiasts to a tool everyone uses daily — and MCP integration is the entrance to that.

A real example combining AI and traditional automation

Let's see, with a concrete example, what combining AI integration and traditional automation makes possible. Suppose someone has a weekly task of aggregating Excel files arriving from various clients. The file formats differ slightly by client, so the same macro can't handle them all each time — yet preparing a macro per client is a lot of work.

This is where AI integration helps. Ask the AI "aggregate the Excel files in this folder into the usual report," and it checks the contents, absorbs the format differences, and assembles a process to fit the moment. And for the routine part after aggregation — "output the report in the fixed format" — the AI calls a macro you prepared and runs it. The ideal division is realised: the AI flexibly handles the changing parts, and a reliable macro the fixed parts.

The strength of this combination is that it fits perfectly the "a bit different each time but the same at root" work common in practice. Work you kept doing by hand because it couldn't be fully templated becomes automatable by combining the AI's flexibility with the macro's reliability. AI integration opens new possibilities in the domain long considered "hard to automate."

Of course, even such advanced integration needn't aim for perfection at once. First prepare basic macros, then make them callable via AI integration, and gradually widen "the range you leave to the AI." With this staged approach, you can make the new automation of collaborating with AI your own — comfortably yet steadily.

MCP and the future of automation

Now that AI assistants can operate a PC directly, automation is shifting from "something experts build" to "something anyone can use through conversation." That said, even when you leave things to the AI, you still need a reliable foundation of automation behind it — recorded macros and verified scripts — because the AI is the conductor while the foundation does the actual work.

By combining GUI recording, SCR scripting and AI integration via MCP in one app, makuroku lets even individuals start this "AI conducts, a solid foundation acts" automation for free. Connect MCP integration on the free plan and ask the AI to "do my usual task." You'll experience your own PC starting to move through conversation alone.