- Automate the fetching and the counting, not the judgement.
- An automated step that cannot be checked is a step you redo by hand.
- Price every run before it starts, or the automation spends on its own.
- Recurring watches turn one research question into a standing one.
What is worth automating
Most marketplace research workflows are the same four steps. Find the listings. Open each one. Copy the price, rating and reviews into a sheet. Then count what recurs. The first three are tedious, and the fourth is arithmetic.
Those are what the agent does. What is left for you is the part that needs judgement: whether the gap is real, and whether it is yours to fill.
- Finding listings: a category scan, by search query or pinned ids.
- Opening each one: price, star histogram, specs and reviews.
- Counting: price spread, 1–2 star share, complaints by brand.
Automation that shows its work
Automated product research fails quietly when it hides its sources. You get a confident summary, it is wrong about one listing, and nothing tells you which. So the research gets redone by hand.
Here the agent shows each tool call as it runs, and every number in the answer links to the listing it came from. The insights card draws the model's read beside the listings. A claim can be clicked and checked against the review behind it.
Priced before it runs
An automation that spends without asking is a bill waiting to happen. Every tool states its credit price before it runs. A category scan is 3 credits per listing and confirms the total first.
Each run and its cost are listed under Usage. A failed call is refunded automatically, and polling a running scan is free.
Tip:Start with the default ten listings. Widen to fifty only once the first ten say the category is worth it.
From one question to a standing one
Once a question is worth asking twice, make it a watch. Recurring watches re-run on your cadence and send only what changed to email, Slack or Linear.
Teams that already run their own agents can skip the web app entirely. The same tools are an MCP server, so a scheduled agent can call them directly.
Frequently Asked Questions
Can I automate Amazon product research with Nooticr?
Yes. Ask a question about a category and the agent scans the listings, reads their reviews and returns the rollup. From your own agent, call scan_amazon_category over MCP.
Does the automation make decisions for me?
No. It collects, counts and drafts a read with every source attached. Deciding what to sell or charge stays with you.
How long does an automated category scan take?
Minutes for ten listings, because each is a real browser render. The scan returns what is ready and finishes in the background.