Introduction
A while back, I wrote a post about switching my main tool from ChatGPT to Claude. It's only been a few months since I made Claude my primary AI, but I've been really happy with what it can do, and I've grown quite fond of it.
I'm still in the trial-and-error stage, but recently I got curious about Claude's "Connector" feature and started experimenting with it. As a simple first step, I connected my Notion and Google Drive — and the experience changed my relationship with AI more than I expected. So I figured I'd jot down a few notes, partly as a memo to myself.
This is fairly introductory content, so it may not be useful for those who are already familiar. But if you're anything like the past version of me — someone who has heard of MCP but doesn't quite have a feel for it — this might be worth a read. Feel free to stick around if you're curious.
What is MCP?
In short, it's a common standard for connecting AI to external tools and data.
Here's an analogy. Imagine you have an SSD full of data. As long as it supports the USB-C standard, you can plug it into just about any device with a single cable and access the data. MCP plays a similar role: as long as a service supports it, Claude can read directly from Notion or reference files in Google Drive.
From "Asking" to "Delegating"
This is just my personal take, but until now, whenever I asked AI for something, the process went like this: organize the situation in my head, summarize it or copy-paste the relevant info, and then ask the question. Once I connected MCP, the AI started fetching the context on its own.
It sounds like a subtle change, but the difference in feel is pretty significant. I had actually known about the concept of MCP for a while, but I'd dismissed it with "if I need to reference something, I can just paste it in and explain, right?" Once I actually tried it, though, the back-and-forth of information became dramatically faster — and as a result, all the work that I should be moving forward (brainstorming, planning, tasks) sped up significantly.
Also, at some point — I don't remember exactly when — Claude gained the ability to reference past conversations (currently only on the Pro plan). With most of the upfront context-setting no longer needed, the pace of pretty much everything went up another notch.
I get a sense that this is a step beyond what AI used to be — not just a "tool to talk to," but more like an assistant or a partner who works alongside you. It might be an overstatement, but that "wait, this is different from before" feeling I had when I first started using it has stuck with me even now.
What I've Actually Tried
Just a few concrete examples. Here are some of the ways I've been using it:
Weekly reviews: Once a week, I keep a record of the previous week's data (logs of what I did, daily notes, work memos, and so on). Before, I would manually go through them every Sunday, clean up notes, and summarize. Now I just point Claude to the right database and ask, "Read through this week's journal and summarize it," and it's done. The whole copy-paste-and-organize step disappeared. Of course, hallucinations and AI mistakes do happen, so it's safer to look over important parts yourself — but those cases are rare, and even when verification is needed, it's still far more efficient than doing everything by hand.
Digging up past decisions: When I ask things like "I remember deciding something like this a while back — is the rationale recorded anywhere?", it'll search for related pages and pull them up. The task of searching through my own Notion has, in a sense, been handed off to AI.
Data analysis: I often keep work management sheets and data in Google Sheets. This kind of analysis is right in AI's wheelhouse — depending on the data, it's quite handy for spotting trends or thinking through strategy.
None of these are flashy use cases, but the small daily frictions slowly disappearing adds up in a way that's quietly powerful.
About Permissions
For now, I'm a bit nervous about AI accidentally deleting or editing important notes, so I'm running everything in read-only mode with no edit permissions.
Depending on what I'm working with and which tools I connect down the line, I might grant edit access later — but even with read-only, it's genuinely useful.
Conclusion
I've only just started, so I'm still getting a vague sense of how useful this really is. I'll keep experimenting, and if I gather enough material, I might write a follow-up.
Thanks for reading to the end.
Top comments (0)