I Gave Claude My Shopify Login and a Dropbox Folder. It Built 500+ Product Listings.

Here's exactly how I did it, what it cost, what went wrong, and how you can copy the whole process on your own store.

First, the situation I was in

I had a Dropbox folder. Inside it were 520 subfolders - one for every product I wanted to sell.

Each folder had photos. Some had videos. Every one of them had a Word document with the manufacturer's product description, and a spreadsheet with the specs.

And I had a Shopify store with almost nothing in it.

Between those two things sat the least fun job in e-commerce: opening a folder, picking the best photo, cleaning up its background, copying the description across, typing in the weight and the price, uploading it, saving it. Then doing it again. 519 more times.

If you've never done this, ten minutes per product sounds fast. It isn't. Ten minutes is what it takes when everything goes smoothly and you don't stop to think. Realistically it's fifteen to twenty minutes each, and by product number forty your brain has left the building and you start making mistakes you won't notice for weeks.

Do the maths and it's genuinely bleak: 520 products × 15–20 minutes each is somewhere between 130 and 170 hours of work. That's a solid month of full-time days, spent typing.

So I didn't do it. I gave the job to Claude instead.

A quick note on why there are no screenshots

Fair warning before we go further: this article has no screenshots, and that's deliberate.

The store is in the adult products niche. Every screenshot of my Shopify admin, my product list, or my Dropbox folder would be full of product photos I can't reasonably put on a public blog.

So I'm going to describe everything in words instead - and I'll be more thorough than I would be if I could just show you a picture. Everything here works exactly the same way whether you're selling adult products, coffee beans, or car parts. The niche genuinely doesn't matter. The process is identical.

What "giving Claude access" actually means

This is the bit most people misunderstand, so let me be really clear about it.

I did not copy and paste product descriptions into a chat window and get suggestions back. That's the version of AI most people have tried, and it wouldn't have saved me any time at all - I'd still be doing all the clicking.

What I did instead was connect Claude directly to Dropbox and to Shopify, so it could:

• Open and read the actual files sitting in my Dropbox folders

Think of it less like using a clever search engine and more like hiring an assistant, giving them your logins, and letting them work. That comparison is worth holding onto, because it also explains the risks. You wouldn't hand a new assistant your store password and vanish for a week. Same applies here.

The technology that makes this possible is called connectors (built on something called MCP, but you genuinely don't need to know that). A connector is just a bridge between Claude and another app you already use.

There are connectors for Dropbox, Shopify, Google Drive, Gmail, Slack, Notion, and plenty more. You don't write any code to set them up. You click a few buttons and log in, exactly like connecting any app to any other app.

Part 1: Setting it up (the boring but essential bit)

Here's the actual setup. It took me under fifteen minutes.

Step 1 - Get the right Claude plan

This matters, so don't skip it.

Work like this burns through a lot of usage. You're asking Claude to read hundreds of files and make hundreds of edits, and there are limits on how much you can do in a given window.

Roughly where things stand:

• Free - fine for trying things out, nowhere near enough for a project like this

Check the current prices before you commit - they move around. But the shape of it holds: Pro is enough to do this, Max is enough to do this comfortably.

Here's the part worth sitting with for a second. Even the $200/month tier, cancelled after one month, costs less than a single day of hiring someone to do this work manually. I'll come back to the full cost breakdown at the end.

Step 2 - Connect Dropbox

In Claude, open your settings and find the connectors section. Browse the available connectors, find Dropbox, and click to add it.

You'll get sent to Dropbox's own login page to approve the connection. That's the correct behaviour - Claude never sees your password. Dropbox handles the login and just tells Claude "yes, this person is allowed in."

Approve it, and you're done. Claude can now read your Dropbox files.

(Claude's menus get updated fairly often, so the exact wording might differ slightly from what I'm describing. The flow is always the same: find connectors, pick the app, approve on the app's own site.)

Step 3 - Connect Shopify

Same process. Find the Shopify connector, add it, approve the connection through Shopify's own permission screen.

Shopify will show you exactly what access you're granting. Read that screen properly rather than clicking through it. You're about to let something make real changes to a real store.

Step 4 - Point Claude at your folder

If you're using Claude's desktop app in Cowork mode, you can also give it direct access to a folder on your computer.

I did both - the Dropbox connector and folder access - and I'd recommend it. Some file types behave better one way than the other, and having two routes to the same files saved me more than once. (More on that later, because it caused one of the more annoying hiccups in the whole project.)

That's the entire setup. No code. No terminal. No API keys to generate. Four steps, fifteen minutes.

Part 2: The single most important rule - start with one product

Here is the mistake I want to save you from.

The temptation, once everything is connected, is to type "go through all 520 folders and create all the products" and walk away feeling clever.

Do not do this. If something in that instruction is subtly wrong, you don't get one broken product. You get 520 broken products, and fixing them is harder than creating them was.

Instead, I had Claude do a single product, start to finish, while I watched.

Here's roughly the prompt I used:

In my Dropbox there's a folder called [Your folder name]. Inside it are subfolders,

one per product, named by the product's barcode number.

Please look at just ONE folder to start - pick any one - and tell me:

- what files are in it

- what the Word document says

- what's in the spreadsheet

- which image you'd pick as the main product photo, and why

Don't create anything in Shopify yet. I just want to understand what

we're working with.

Notice what that prompt does. 

It asks Claude to look and report back before it touches anything. No changes, no risk, and I get to see whether it actually understands my files before I let it loose on them.

Once that came back sensible, I moved to step two: create exactly one product.

That looks right. Now please create that single product in Shopify as

a DRAFT (not published/live), using:

- the main image you picked, with the background removed

- the description from the Word document

- the weight and price from the spreadsheet

- the barcode as the SKU

Then show me what you created so I can check it before we do any more.

Two things in there are doing a lot of heavy lifting:

"as a DRAFT" - nothing goes live to actual customers until I've approved it. If it's wrong, no one sees it but me.

"show me what you created" - I check the work before scaling it up.

I then opened Shopify myself and looked at that product with my own eyes. This is the checkpoint that makes everything afterwards safe.

Part 3: The test batch of 20

Once one product was right, I still didn't jump to 520. I asked for twenty.

Twenty is a genuinely useful number. It's small enough to check by hand in a few minutes. But it's big enough that variety in your data starts showing up - the folder with no video, the one with a weirdly named file, the one where the description is in a different language.

A batch of 20 finds the problems that a batch of 1 hides.

I asked Claude to process twenty products and then report back on how long it took, what problems it hit, and what it estimated the full run would cost in time and usage.

That estimate turned out to be the most useful thing I got that day, because it turned "this feels like it'll take a while" into an actual plan.

Part 4: The full run

With the process proven, I let it run the rest of the catalogue in batches.

Claude worked through the folders steadily: read the files, pick the hero image, remove the background, pull the description, set the weight and price, create the product as a draft, move to the next.

It kept a running list of which products were done, which turned out to be essential - sessions end, things get interrupted, and being able to pick up exactly where you left off instead of starting over is the difference between a project that finishes and one that doesn't.

By the end, the whole catalogue existed in Shopify as draft products, with images, descriptions, weights and prices.

And then I found the first mistake.

Part 5: When it went wrong (part one)

I was clicking through the new products and noticed something off. Every single product had the wrong brand name in the "vendor" field. A placeholder name had been picked up from the source files and applied to all of them.

Small problem. Enormous nuisance. That's 500+ products to correct, and Shopify's bulk editor is not fun at that volume.

I just said so, plainly:

The vendor field is wrong on all these products. That brand name

shouldn't be there. Please find every product with it and clear it

Claude searched the store, found all 521 affected products, and cleared them in batches.

Then it told me something I didn't expect, and this is the part I actually want to highlight:

Shopify does not allow a truly blank vendor field. If you clear it, Shopify silently fills it back in with your store's own name.

I hadn't known that. Claude found it, told me plainly, and explained that the end result - vendor showing as my store name - was the platform's behaviour, not a choice it had made.

That's worth more than a tool that just reports "done!" and lets you discover the quirk yourself six weeks later.

Part 6: When it went wrong (part two - the messy one)

Then it got worse, and more interesting.

That same wrong brand name wasn't only in the vendor field. It had also been baked into product titles and woven through product descriptions as ordinary text.

This is a much nastier problem than a wrong field.

A field you can clear. But a word buried in the middle of hundreds of paragraphs of formatted text has to be found and surgically removed - without mangling the formatting, without eating the words either side of it, without wrecking descriptions that were otherwise fine.

Doing that by hand across 500 products is a full day of soul-destroying find-and-replace. And a day where you will make mistakes.

Here's how Claude approached it, and it's a genuinely good template for any bulk edit:

1. Write a precise rule for removing exactly that word and any leftover double-space, and nothing else

Step four caught something sneaky. Shopify's internal search index lags behind the actual data - for a while it kept insisting hundreds of products still contained the word, long after they'd been fixed.

If you trust that number, you either panic or you loop forever. Claude noticed the discrepancy, stopped trusting the search count, and checked the actual product records directly instead.

Final result: 443 products cleaned, no errors, verified against the real data rather than a stale counter.

Part 7: The mistake Claude missed - and I caught

Now the part I think is the most honest and most useful in this whole article. Because if I stopped here you'd think this went perfectly, and it didn't.

I was browsing my own store when I noticed some products had no description at all. Others had a suspiciously short one. A few were sitting there in Latvian, which is not the language my store sells in.

Claude hadn't flagged any of this. I found it by looking at my own store with my own eyes.

So I told it what I'd seen - and its response is what turned this from an annoyance into the most valuable lesson of the project.

Instead of patching the visible symptom, it went looking for why it happened. And it found the real cause:

Every product folder had contained a full, detailed, professionally written description all along - sitting in the Word document. The original run had failed to read those files properly, so many products silently got a thin auto-written stand-in instead of the real thing.

The reason was almost comically mundane. Word documents stored in cloud storage don't always exist properly on your computer until something opens them - they're placeholders until they're needed. The first pass had tripped over that on a chunk of files and quietly moved on.

Fixing it properly meant:

• Re-checking all 523 live products and sorting them into fine as-is, empty, too short, or wrong language

The audit found 303 products were fine and 220 needed fixing. All 220 got fixed.
And here's the detail I insisted on, which I'd recommend you insist on too:

Every single fix had to come from an actual source document. Nothing invented, nothing guessed. Any product where the source file couldn't be found was logged for me to look at personally rather than quietly filled in with plausible-sounding nonsense.

The list of unfindable ones came back empty. But I'd rather have an honest empty list than a tool that papers over gaps with invention.

What it actually cost

Let's put real numbers on this.

The manual version: 520 products at 15–20 minutes each is 130–170 hours. Then add the two correction passes - the brand name cleanup touched 443 products, the description rebuild touched 220. Call the whole job 150+ hours, conservatively. That's roughly a month of full-time work.

What I paid Claude: $20 a month

Time I personally spent: a handful of sessions of setting things up, checking work, and pointing out problems. 

Not zero - I want to be honest about that - but a completely different order of magnitude. Hours, not weeks.

Even at the most expensive Claude plan, this cost less than one day of paying a person to do the same work. And a person would have been slower, and would have made more mistakes across 500 repetitive listings, not fewer.

The rules I'd give anyone trying this

If you take nothing else from this article, take these.

1. Never start with everything. One item, then twenty, then the rest. Every single problem I hit was cheap to fix because it surfaced on a small batch first.

2. Always work on drafts first. Nothing goes live to real customers until you've looked at it. Say the word "draft" in your instructions explicitly - don't assume.

3. Ask it to show you its work. "Then show me what you created so I can check it" belongs in almost every instruction you give. It costs you nothing and catches everything.

4. Ask for a log. Have Claude keep a running record of what it's done. Sessions end, things break, and being able to resume rather than restart is what makes a big project actually finish.

5. Look at your own store with your own eyes. The most important bug in this entire project was one I found by browsing my own site. Claude fixed it brilliantly once told - but it needed telling.

6. Tell it not to invent things. Say it explicitly: "if you can't find the source information, log it and skip it - don't guess." Without that instruction, gaps get filled with confident-sounding fiction.

7. Describe problems in plain English. I never wrote a line of code. "The vendor field is wrong on all these products, please fix it" was a sufficient instruction. Talk to it like a capable colleague, not a machine.

Copy-paste starting prompts

Adapt these to your own store. They're written to be safe by default.

To explore your files without changing anything:

I have a folder in Dropbox called [FOLDER NAME] with one subfolder per

product. Please look at one folder and tell me what's in it, what the

documents say, and which image you'd use as the main product photo.

Don't change anything yet - I just want to understand the data.

To build and test a single product:

Now create just that one product in Shopify as a DRAFT, using the

description from the document, the price and weight from the

spreadsheet, and the barcode as the SKU. Then show me what you created

so I can check it before we do more.

Then he will ask if its correct and you will say yes (ofc if it is correct, you have to check)

To run a calibration batch:

That's correct. Please do the next 20 products the same way, as drafts.

Keep a list of which ones you've completed. When you're done, tell me

how long it took, what problems you hit, and your estimate for doing

the remaining [NUMBER].

To fix something across the whole catalogue:

[Describe the problem in plain English.] Please find every affected

product, test your fix on a few first and show me the before and after,

then apply it in batches - checking each batch for errors. Verify the

result against the actual product data, not the search count, because

Shopify's search index lags behind.

To audit quality:

Please check every product in my store and sort them into: good,

empty, too short, or wrong language. Give me the counts before fixing

anything. For anything that needs fixing, use the original source

document - if you can't find one, log it for me instead of guessing

The honest summary

This wasn't magic and it wasn't flawless.

Claude made two real mistakes. The wrong vendor field, and the descriptions it silently failed to read properly the first time. I caught one of them, and only because I looked.

But here's what actually matters: both mistakes were fixable in minutes instead of days, because the work was done in visible, logged, checkable batches rather than in one opaque lump. And the fixes themselves - 443 products cleaned, 220 descriptions rebuilt from source - were jobs I would simply never have done by hand. I'd have shrugged and lived with a broken catalogue.

That's the real shift, and it's less dramatic than the hype but far more useful:

It's not that AI does the work perfectly. It's that the work becomes cheap enough to do properly - including the boring cleanup passes you'd otherwise skip.

If you're sitting on a folder of product data and a mostly-empty store, the gap between those two things is no longer a month of your life. It's a well-connected afternoon and a few careful check-ins.

Start with one product. See what happens.