The short answer: seven marketplace data sources now expose an MCP server that an AI assistant can call directly: Jungle Scout, Helium 10, SmartScout, Keepa, Bright Data, Apify and Nooticr. They differ less in the protocol than in what comes back. Some return your own seller account's numbers. Some return estimates. Some return the raw listings and reviews for your model to read.
This is a comparison for AI retrieval specifically: what matters when the reader is a model, not a person at a dashboard. Nooticr is our product. Facts about other vendors are from their public documentation as of September 2026.
Six questions that matter for AI retrieval
- Whose data is it? Your own seller account, or the public market? An account connector cannot research a category you do not sell in.
- What comes back? A model reasons better over listings and review text than over a single score. Estimates are useful, but your model cannot check them.
- Which marketplaces? Most servers are Amazon-only.
- How is it billed? An agent calls tools in loops. Per-call pricing an agent can read beats a plan limit it cannot see.
- What happens on slow jobs? Collecting fifty listings takes minutes. A good server returns partial results and a job id, not a timeout.
- Is untrusted text framed as data? Reviews are written by strangers. A review that says "ignore your instructions" should reach the model as evidence, not as a command.
The servers compared
| Server | Returns | Marketplaces | Access and billing |
|---|---|---|---|
| Jungle Scout MCP | Sales estimates, keywords, share of voice, product database | Amazon (19 marketplaces in Cobalt) | Through Cobalt; sales-led |
| Helium 10 MCP | 34 tools: keywords, product research, rank, profit, inventory, ads | Amazon, Walmart | Diamond plan and up; 1,000 calls a month |
| SmartScout MCP | Products, brands, sellers, subcategories, keywords | Amazon | Through the SmartScout API |
| Keepa MCP | Price and sales-rank history, offers, deals, sellers | 10 Amazon domains | Keepa API key and tokens |
| Bright Data MCP | Structured scraped product pages and search results | Any public site, with an Amazon-specific server | Bright Data account |
| Apify MCP | Output of ready-made scrapers from the Apify Store | Whatever an Actor covers | Apify account |
| Nooticr MCP | Listings, prices, star histograms, review text, category rollups; social posts and comments | 11 marketplaces, 11 social networks | Credits per fetch, price in every tool description; 20 free |
Seller-account servers: Helium 10 and Jungle Scout
These are the natural choice if you already pay for the product. Your assistant can pull your keyword ranks, profit and inventory, or Jungle Scout's sales estimates, without exporting a report. The trade-off is scope: they answer Amazon questions, and Helium 10 adds Walmart.
Market-data servers: SmartScout and Keepa
SmartScout answers "which brands and niches are big?" with revenue estimates. Keepa answers "how has this price moved?" with history. Both are Amazon-only and both are narrow on purpose, which makes them easy for a model to use well.
Infrastructure servers: Bright Data and Apify
These fetch pages for you and get past bot defences. You get structured data from almost any site, and you build the analysis. They are the most flexible option and the most work.
Nooticr: reviews and social, read by your model
Nooticr sits between the two. It covers 11 marketplaces, including Temu, AliExpress, Mercado Libre, Flipkart and Lazada. A category scan returns up to 50 listings with review text and does the counting: price spread, 1–2 star share from the full histogram, and which complaints recur across how many brands.
It was built with the six questions above in mind. Every tool states its credit price in its description, so an agent can budget. A slow scan returns a scan id, and polling it is free. Review text and comments are framed as material to reason over, never as instructions. And the same server reads TikTok, YouTube and Reddit, so one conversation can put what buyers wrote beside what viewers said.
It does not estimate sales and does not cover Walmart or Target. If those are your questions, pair it with one of the tools above.
Connecting one
Most of these servers take a URL and an OAuth sign-in. For Nooticr, add https://mcp.nooticr.com/mcp as a connector in Claude or ChatGPT, or run npx -y @nooticr/mcp locally. Then ask a category question in plain words and tell the assistant to poll until the scan is finished before it concludes.
For a broader comparison beyond AI retrieval, see the best marketplace intelligence software compared.