The analyst, not the dashboard.
Claude reads the live market across sources for your question, weighs every signal for and against, and gives a verdict for you, not just for the market, with the one next move. It all runs in your own Apify account, every row is checked before it counts, and the verdict is read off published rules. This page is the method, in full.
Claude reads the live market across sources, weighs every signal for and against, and gives a verdict for you, with the one next move.
Six steps, every time.
You ask, in plain words
"Is this worth selling?", "Who sells below my minimum price?", "What does rival.com run on sale?" Claude picks the right kind of research; you never have to name one. Not sure where to start?
/marketmage:startshows seven things you can ask, with an example each, and asks what brings you here.A yes before anything paid
Small reads just run. A bigger run, or anything that repeats, gets one estimate first, with the free-plan price too where it differs, and waits for your yes. That's the only place it talks about cost.
It runs in your account
CleanPull's data tools run in your own Apify account. Every run has a spending limit a little above its estimate, never above a budget you set. The results are saved in your account, not ours.
Every row is read
Before any figure reaches you, each result is checked: is it the same product, the same condition, the same pack size? What fails is set aside, and the answer says how many.
Your answer first, then the verdict
The reply opens with a plain answer to what you asked. Then the verdict: each signal checked against a published threshold, and Strong, Promising (with one fix) or Not this version read off a fixed rule, never a mood. Share how you sell, and a verdict for you comes first. It ends on the one next move.
Any question can become a watch
Say "watch this" and the same research runs on a schedule, with a cost limit per run, and sends you only what changed.
Sources read together, by how you sell.
No single source settles an idea. TikTok Shop sales show people are buying; Meta ads still running after 35 days show a seller it keeps paying for; Google Trends shows the timing; supplier prices show the margin. So when you ask for ideas, Claude reads the sources your channel's buyers act on, and an idea that several of them back ranks higher, with its backing said in plain words.
Claude reads the sources your channel's buyers act on. An idea several of them back ranks higher, with its backing said in plain words.
Dropshipping, or your own store with adsTikTok Shop, Meta ads running 35+ days, AliExpress single units and Google Trends.
TikTok Shop units sold and the ads that have lasted come first; then, for the product types they surface, whether AliExpress sells one unit at a time and at what price, and the search trend. The AliExpress price is what a repeat buyer pays, with no sign-in; first-order "New shoppers" deals are kept separate, so they never flatter a margin. Shipping to your buyer's country is added from the product page for the landed cost.
Selling on AmazonBest Sellers charts, how crowded each search is, and Google Trends.
The charts show what sells now; the competition check shows how deep the leaders' reviews go and how concentrated page one is; Trends shows whether interest is rising.
Selling on Etsy or eBayThat marketplace's own listings, and Google Trends.
Prices, sellers and review depth from the listings on the marketplace you sell on, with the trend beside them. Amazon's page one is context for you here, never counted against an idea.
A source that wasn't read never counts against an idea, and you can add or skip one: "also check TikTok", "skip Amazon".
Illustrative, shortened. Ideas one source shows say so ("TikTok only so far").
The right product, like for like. Then the numbers.
Marketplace searches are messy. A "lip gloss" search brings back the brand's deodorant and serum, an Amazon top ten includes accessories, a price list mixes open-box units with new ones. Claude checks each row before it counts, and tells you what it left out.
Marketplace searches are messy: accessories in a top ten, open-box units among new ones. Claude checks each row before it counts, and tells you what it left out.
Same productAn accessory, a bundle or another product of the same brand isn't the product.
Claude reads the titles instead of matching keywords, and treats the data tools' own relevance flags as clues, not verdicts.
Like for likeA renewed or open-box offer never counts against a new price.
A multi-pack priced as one unit is set aside before prices are compared. Where the condition isn't stated, it's reported as unknown, never as new.
Says how many"2 of the 10 leaders were accessories." "13 of 48 listings were renewed."
If too little is left to judge a signal, that signal is marked unknown rather than guessed.
- Earbuds, noise-cancellingcounted
- Replacement ear tipsaccessory
- Earbuds, sport fitcounted
- Earbuds, renewedrenewed
- Earbuds, with casecounted
- Charging case onlyaccessory
- Earbuds, two pairs as one2-pack
- Three more pairscounted
Judged on the 6 new, single pairs. 2 of the 10 leaders were accessories, 1 was renewed and 1 was a two-pack priced as one, so those were set aside.
Always the same shape, whoever asks.
Claude scores each signal first, with the number and its threshold side by side, and only then writes a verdict, using a fixed rule. The same data gets the same verdict whether you asked "is it worth selling?" or "is it saturated?", and you get one line on the opposite question too.
Claude scores each signal against its threshold first, then writes the verdict by a fixed rule. "Is it worth selling?" and "is it saturated?" get the same verdict, with one line on the opposite question.
The rule
- Strong
- Margin not against, and either five or more signals for with at most one against, or four or more for with none against.
- Not this version
- Four or more against; or margin and concentration both against; or two or more against that outnumber the for and unread signals together. It always comes with a version the data favours.
- Promising
- Everything else, with the one fix. Three or more unknown signals cap the verdict at Promising, with low confidence.
The wordsHopeful and plain. Warm only where a number backs it, and never an adjective about you.
"The data favours this", "the data is mixed: here's the one fix", "the data is against this version; here's one it favours". A real opening is called one, with its number. Never "hot", "winning", "goldmine" or "definitely", and never doom: a weak version is said once, kindly, then the answer builds.
The other sideEvery verdict carries its counterweight.
A Strong verdict names what would need to stay true and the early sign it isn't. Not this version names related versions the data favours, and offers the best one for a check. Promising names the one fix and the test that settles it.
The rest of the answer5 more parts
Illustrative figures, in the plugin's own answer format, as an experienced seller sees it. A first-time seller gets the same verdict in plain words: counts said in words, a table that says helps, hurts or neutral with a short reason, and the thresholds in the full detail. A first answer has no chart; the full detail, and a chart where Claude can draw one, come when you ask.
Confidence is coverage, not feeling.
Every verdict states its confidence, and it means one thing: how much of the planned data came back.
- High
Every planned source came back, with at least its usual number of results.
- Medium
One planned source is missing or thin: under 20 reviews, under 20 listings, or a weak Google Trends line.
- Low
Two or more sources missing or thin, or three or more signals unknown. The answer says which source would raise it.
Every answer names its blind spots.
A source that isn't planned for your marketplace isn't "missing". TikTok Shop covers the US shop only, so outside the US it simply isn't read, and the answer says so once.
The limits of every source are listed on the coverage page, and repeated, one line a source, in the full detail of every answer that used them.
"For anyone", apart from "for you".
If you like, tell Claude roughly what you could put into a first order, how you ship, whether you've sold before, where you sell and whether this is a side income or a brand. The facts and the market verdict never change with it. A separate, labelled "For you" block does.
If you like, tell Claude your budget, how you ship, your experience, where you sell and your goal. The facts and the verdict never change with it. A separate, labelled "For you" block does.
Your budget, kept as a range
- Under $1k
- $1k to $5k
- $5k to $25k
- More
Ranges, never figuresAn exact figure you mention is turned into a range, and never repeated or stored.
Along with the budget range, Claude notes how you ship, your experience, where you sell and your goal.
It stays in the chatUnless you turn on the notebook, it's never sent to Apify or saved anywhere.
If you want it remembered, say yes to the optional notebook: Claude saves your profile there as bands only (never amounts), in your own Apify account, where you can view or delete it.
The market verdict never movesYour profile changes only the "For you" verdict, never the facts or the market's.
The "For you" verdict reads the same facts through how you sell: the checks that matter to you (a dropshipper's ad-policy and single-unit checks, an FBA seller's first order) and a fit check with numbers. When the two verdicts differ, Claude says why in one sentence.
When a quick answer isn't enough, go deeper.
After an answer, Claude offers to go deeper and shape it around what you'd actually sell. It's a conversation that narrows, not a longer report: your picture of the buyer, two or three directions built from facts already read, then one thing at a time until you have a shortlist and a concrete next step.
After an answer, Claude offers to go deeper: your picture of the buyer, two or three directions built from facts already read, then one thing at a time until you have a shortlist.
Your vision firstOne or two sharp questions, never a form.
Who you picture buying it, what you'd rule out, what you've already tried and what went wrong. What you've already said is never asked again. The estimate comes in the same message, so answering and saying go is one turn.
Directions, not verdictsEach one starts from a fact read in the chat.
A rising search, a TikTok listing, a leader's complaints. Each direction is a short card: who it's for, its demand, the gap with its count, whether a factory can make it at your size, and the open question samples must settle.
Your complaints, found in reviewsThe problems you name are looked for first.
Claude searches the leaders' low reviews for them and gives the count. A gap backed by only a few reviews is called a hint, not a finding.
From a real answer, November 2026, shortened. US review text needs a sign-in, so the same leaders were read on Amazon UK.
Answers that fit how you read.
The facts and the verdict are the same for everyone. How they're said follows you.
Plain words by defaultEvery trade term explained the first time it comes up.
A first-time seller gets sentences, counts in words and a table that says helps, hurts or neutral with a short reason. Experienced sellers, or anyone who uses the trade terms, get the compact form with each threshold beside its number.
In your languageWrite in Spanish, German, French, Italian or Portuguese.
The answer comes back in your language, with the verdict labels in fixed translations, the same words every time: "Para ti: este no · Para el mercado: Esta versión no: esta otra podría funcionar".
Read aloudUse a screen reader? Say so.
No tables and no symbols: every fact becomes a labelled line ("Search trend: helps, up 38% over five years"). With the notebook on, the layout is kept for next time.
"Too late?" gets a straight answerEarly, growing, crowded at the top, or cooling.
Read from the demand trend, how deep the leaders' reviews go, and whether newcomers in the top 10 are getting sales. There's no "too late" label: each reading says where the room is, with the two facts that decide it and what a sale would leave if ten more sellers arrive.
A line before you orderFor products that are often regulated.
LED masks, supplements, hearing amplifiers and the like get one plain line on what to check first: the supplier model's FDA status, Amazon's approval for the category, and the claims a listing can't make. It never changes the verdict on its own.
Picks up where you left offWith the notebook on, "where were we?" just works.
Claude reads your notebook without asking, welcomes you back at your stage and offers its next step. If you have watches, your briefing adds what they found.
Any question can keep asking itself.
A watch is saved in your Apify account, where you can see and change it. The first run records a starting point; from the next, you get one digest of only what changed.
- Prices
- Sales
- Stock
- Who wins the sale (the Buy Box)
- New products
- Amazon ranks
- New reviews
- Promotions and launches
- New sellers in your searches
- A competitor's new ads
Alerts narrow itOnly what crosses the line you set.
Only reviews below a star rating you choose, only price drops of at least a percentage you choose, or only when a rank passes a milestone.
In your Claude briefingAsk "catch me up", or have Claude brief you every Monday.
Every digest is kept in your Apify account. Your briefing reads them for free and turns what changed into a few decisions, when you ask or on a Monday schedule (Claude's own scheduled tasks, on paid Claude plans). Slack, Discord, Teams or email only if you ask for one; it's added on the watch's page in the Apify Console, never in the chat.
Only real changesA page that couldn't be read in full is never reported as a change.
Stock that couldn't be read stays unknown, never "sold out", and every digest says what was checked.
Managed by asking"What am I watching?" "Pause my rank watch." "What does it cost?"
Claude pauses or changes a watch for you. Deleting one for good is done in the Apify Console.
One price change across 12 stores
Noise-cancelling headphones
- AmazonDown from $198.00 to $189.99, now the lowest new price.
- WalmartBack in stock.
- New storeUnder $200, but open-box, so it was set aside.
Your other stores: no price change.
Your yes before anything paid. Then a limit on every run.
Installing it costs nothing; runs bill your own Apify account at the tools' published prices. Three things keep a run from costing more than you expected.
One estimate, at the yes
Above 5 cents, or before anything that repeats, Claude gives one estimate from the published prices ("about $0.20, occasionally a little more if I need to dig deeper to answer you") and waits for your yes; under that, a read just runs. No cost lines on answers or next steps.
A limit on every run
Every run started in the chat has a spending limit a little above its estimate, never above a budget you set, so a job can finish without asking twice and still can't run away. A watch has a maximum cost per run, which every CleanPull tool keeps to; scheduled exports have a row limit instead.
Secrets never in the chat
The only secrets are your own delivery settings. Claude links you to the watch's page in the Apify Console; you paste them there, and Apify stores them encrypted. Paste one into the chat and Claude refuses it.
Every per-result price, the example monthly costs and how the limits work: pricing.
What it can do in your Apify account. And what it can't.
You connect with Apify's own sign-in (OAuth), and we never see or store a token. Apify gives that sign-in full access to your account. MarketMage's connector narrows what Claude can use to 13 named tools.
The 13 tools Claude can use
- Start a run of a data tool
- Read a run's status, and the list of runs
- Read results and stored records
- Stop a run
- Create, read and update saved tasks
- Create, read and update schedules
- Read the Apify API, with read-only requests
What MarketMage does with them
- Starts only CleanPull's data tools, and reads their results.
- Reads your tasks and schedules to find its own watches, named
cw-…, and creates, changes or pauses only those, when you ask. - Gives every run started in the chat a spending limit, and every watch a maximum cost per run.
What it can't do
- Delete anything: there is no delete tool. You delete a watch in the Apify Console.
- Change your Apify plan, billing or usage limits: none of the 13 tools does that, and the API reader only reads.
- See your payment card: Apify's API doesn't return card details.
Plainly: Which tools it starts and which watches it touches are the skills' instructions to Claude, not a technical lock. The 13 tools could start any Apify actor or read any part of your account, such as runs, results, tasks, schedules and usage. To approve each call yourself, leave Customize → Connectors → apify → Tool permissions on Needs approval. Ask "show me the input" to see exactly what a run was given. More on security and data handling.
Ask like a seller. Be answered like a client.
Add it to Claude today in a few clicks. Once it's in Claude's plugin directory, you'll also find it there.