Marketplace Data API and MCP: 128 E-Commerce Sources for AI Agents

An ecommerce data API returns product, price, seller and review information from online marketplaces as structured records, so you send a product ID, a keyword or a seller and get fields back instead of pages to parse. NeuralVerge's source catalog has 128 such lookups across 24 marketplaces and platforms, reached over REST or by an AI agent through MCP. This guide covers what they return, what they cost and how to judge any ecommerce data API before you build on it.

The short version

• —What you get. 128 catalog entries, each one lookup (a product, a search, a seller, a reviews feed, a category, an offers list), across 24 marketplaces and platforms, all answering in one response envelope. "128" counts lookups, not marketplaces.
• —Who it is for. Teams doing price monitoring, competitor and assortment research, or giving an AI agent live product data. It is not for managing your own listings or reading your own orders; that stays with each marketplace's own seller tools.
• —What it costs. A lookup that resolves one record costs 5 points ($5 per 1,000). A search, list or reviews lookup costs 1 point per returned result ($1 per 1,000). Where a source page says misses are not billed, a request that finds nothing costs nothing.
• —How to reach it. The same lookups are available over REST and as MCP tools, at the same prices. The source catalog is the reference for exact fields, request shape and prices for every lookup.

What is an ecommerce data API?

An ecommerce data API returns product, price, seller, review and category information from online marketplaces and shops as structured data. The caller does not receive a web page. It receives named fields, such as a price, a rating, a review count and an image URL, that a program or a language model can use directly.

That definition is easier to hold on to by separating it from three things it gets mixed up with.

An ecommerce data API vs. a marketplace's own API

Most large marketplaces publish interfaces for their own sellers and partners, designed around running your business there: creating listings, updating stock, reading your orders. They are the right tool for that job and nothing here replaces them.

An ecommerce data API as described in this guide does the opposite job. It reads what any visitor could see: a listing page, a seller storefront, a category, a review feed. It works on listings that are not yours, which is what you need for price comparison, assortment research or competitor monitoring. The tradeoff is that you only get what is publicly visible.

An ecommerce data API vs. a scraper you build

Building your own collection code can be reasonable for one marketplace with stable needs. The cost shows up later: page layouts change, each new marketplace is a new project, and every output looks different. A hosted catalog moves that maintenance behind a request. What you give up is control: if a source lacks a field you need, you cannot patch it yourself.

An ecommerce data API vs. a generic web search or extraction tool

A search tool finds pages. An extraction tool turns a page you already have into fields you describe in a schema, which suits a site that is not in any catalog. A marketplace source is narrower: the input is a product ID, a keyword or a seller, the output has a documented shape, and the price per call is known in advance.

How the catalog works

Every source, whatever the marketplace, is called the same way, which is what lets an agent or a pipeline treat the catalog as one tool instead of 24.

1. Pick the lookup that matches what you have

You start from what is in hand. An ASIN points to Amazon product or Amazon product offers. A keyword points to a search source. A seller name points to a seller lookup. The source catalog lists every source with its input, what it returns and its price, so choosing is a matter of reading one card.

2. Send one request

Each source is a REST function under the same host, called with an API key and a small JSON body. The body carries the lookup input plus any options, such as a marketplace domain or a maximum number of results. The parameter names and defaults differ per source, so the reference for each one lives in the documentation rather than being repeated here.

3. Get back one consistent envelope

The response comes back in the same outer shape for every source. It carries a session ID you can trace, the kind of lookup, the inputs that were used, a human-readable summary, a machine object with the typed records, and the points the call used. The records inside machine differ by source, because an Amazon product and a Poshmark follower are different things, but the envelope does not, so error handling, logging and cost tracking are written once.

What the 128 lookups cover

The sources follow a small number of patterns, which is the main reason a catalog this size stays learnable. Most marketplaces offer some mix of the same lookups: product or item, search, seller or store, reviews and questions, categories, and offers (several sellers listing the same product, as on Amazon and Newegg). A product lookup takes an ID or URL, such as an ASIN; a search takes a keyword; a seller lookup returns a profile and often the seller's products or feedback.

The platform mix is wider than the big retailers: second-hand and classified sites, auction sites, regional marketplaces and Shopify store lookups are all in the tables below. The tables list every marketplace in the e-commerce category, grouped by region, with each lookup under the points it costs. They come from the source pages, counted at the time of writing: 128 lookups in all, across 24 marketplaces.

US and global

Europe

Asia-Pacific

Turkey and Russia

A real request: Amazon product search

Here is one lookup end to end. Amazon product search takes a keyword and returns the listings Amazon's own search page would show for it. It can be pointed at any of 19 Amazon marketplaces with a domain parameter, and defaults to amazon.com. The Amazon product search page is the reference for its full parameters.

The REST request, with the key replaced by a placeholder:

curl -X POST https://api.neuralverge.ai/functions/v1/run-amazon-product-search \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "query": "usb cable", "max_items": 2, "domain": "amazon.com" }'

The JSON below is the captured sample response published on the source page for a request with those inputs, reproduced as it appears there (the session ID is the page's placeholder).

{
  "session_id": "11111111-1111-1111-1111-111111111111",
  "kind": "amazon_product_search",
  "query": "usb cable",
  "domain": "amazon.com",
  "max_items": 2,
  "start_page": 1,
  "partial": false,
  "human": "# Amazon product search\n\n- **Search term:** usb cable\n- **Domain:** amazon.com\n- **Starting page:** 1\n- **Max items:** 2\n- **Records returned:** 2\n- **Proxy traffic:** 572.0 KB",
  "machine": {
    "total_items": 2,
    "first_item": {
      "id": "B07X6CWM3G",
      "title": "6amLifestyle 2 Pack 10ft PS4 Controller Charging Cable,Micro USB Braided Cord Compatible with PlayStation 4 DualShock 4, PS4 Slim/Pro, Xbox One S/X, Android Phones, Kindle, Black & Blue",
      "url": "https://www.amazon.com/sspa/click?ie=UTF8&spc=MToyNzg0MjMyMzk4ODUwMDM3OjE3OTA3ODE3MDQ6c3BfYXRmOjMwMDYxOTg2MjUwNDcwMjo6MDo6&url=%2FPS4-Controller-Charging-Cable-Compatible%2Fdp%2FB07X6CWM3G%2Fref%3Dsr_",
      "price": "$13.99",
      "rating": 4.6,
      "review_count": 23645,
      "image": "https://m.media-amazon.com/images/I/71G1-QuX2IL._AC_UY218_.jpg"
    },
    "items": [
      {
        "id": "B07X6CWM3G",
        "title": "6amLifestyle 2 Pack 10ft PS4 Controller Charging Cable,Micro USB Braided Cord Compatible with PlayStation 4 DualShock 4, PS4 Slim/Pro, Xbox One S/X, Android Phones, Kindle, Black & Blue",
        "url": "https://www.amazon.com/sspa/click?ie=UTF8&spc=MToyNzg0MjMyMzk4ODUwMDM3OjE3OTA3ODE3MDQ6c3BfYXRmOjMwMDYxOTg2MjUwNDcwMjo6MDo6&url=%2FPS4-Controller-Charging-Cable-Compatible%2Fdp%2FB07X6CWM3G%2Fref%3Dsr_",
        "price": "$13.99",
        "rating": 4.6,
        "review_count": 23645,
        "image": "https://m.media-amazon.com/images/I/71G1-QuX2IL._AC_UY218_.jpg"
      },
      {
        "id": "B0DZNBF9NV",
        "title": "ULIFTUS USB-B to USB-C Printer Cable 3FT, Gold-Plated USB Type C to Type B Cord for MacBook Pro/Air HP, USB C MIDI Cable for Yamaha Piano DJ Controller Keyboard, 1m Printer Cord to USBC",
        "url": "https://www.amazon.com/sspa/click?ie=UTF8&spc=MToyNzg0MjMyMzk4ODUwMDM3OjE3OTA3ODE3MDQ6c3BfYXRmOjMwMDc0ODIxMjk0MzgwMjo6MDo6&url=%2FULIFTUS-Printer-Gold-Plated-Controller-Keyboard%2Fdp%2FB0DZNBF9NV%2Fre",
        "price": "$5.99",
        "rating": 4.7,
        "review_count": 478,
        "image": "https://m.media-amazon.com/images/I/71hy5RQ8vOL._AC_UY218_.jpg"
      }
    ],
    "proxy_traffic_bytes": 585701
  },
  "total_points": 2
}

Reading the response

• —kind names the lookup, and query, domain and max_items echo the inputs used.
• —human is a short readable summary for a person scanning logs, and machine is what code should read. The records are in machine.items; first_item repeats the first one for convenience.
• —Each record carries the ASIN as id, the title, the product URL, the price as displayed on the page, the star rating, the review count and an image URL. The price is a string with its currency symbol, not a number, so normalise it before comparing across domains.
• —total_points is the bill for this call: 2, for two returned results.

Note that the sample's url values are search-result click URLs, as they appeared on the results page. For a clean product link, use the ASIN with the Amazon product lookup instead of relying on the search URL.

A worked example: pricing a new product line

This is an illustration, not a case study. Acme Oy, a Finnish company selling charging accessories, wants to understand the market before it lists a new cable on three marketplaces.

1. —An analyst starts with a search lookup for the cable category on Amazon, once on the German domain and once on the UK domain, to see which listings hold the top results and at what price.
2. —For the five listings that appear on both, the analyst runs the product lookup to pull the full details, then the offers lookup to see how many sellers compete on each one.
3. —Two of those sellers look dominant, so a seller lookup and a seller products lookup show what else they sell.
4. —The team repeats the search and product steps on eBay or Walmart to compare channels. Every response has the same envelope, so the same code handles all of it.

REST and MCP

The request above is one REST call. The same lookup is also an MCP tool on the hosted MCP server at https://mcp.neuralverge.ai/mcp. The tool name is the REST function name with underscores (the REST path uses hyphens), and the arguments match the REST body:

{ "tool": "run_amazon_product_search", "arguments": { "query": "usb cable", "max_items": 2, "domain": "amazon.com" } }

Both reach the same sources at the same prices. Every source in the e-commerce category is available this way; AI research is the one NeuralVerge product that runs over REST only.

Use REST for scheduled jobs, pipelines and anything that runs without a person or an agent deciding what to fetch. Use MCP when an agent should choose the lookup itself, for example picking a product search or a seller lookup based on the question it was asked, without a developer scripting every path. Each source page names its MCP tool with a sample call. For connecting a client such as Claude, see the MCP setup guide.

Cost and limits

Every source in the e-commerce category is priced in points, and the source pages state the dollar equivalent. The tables above already split the catalog by price: a lookup that resolves one record costs 5 points, which the source pages state as $5 per 1,000, and a search, list or reviews lookup costs 1 point per returned result, stated as $1 per 1,000.

For the call above, two results at 1 point each is 2 points, which matches total_points in the response. A 25-result search is 25 points. A single Amazon product lookup is 5 points. Where a source page says a missing record is not billed, such as an ASIN that cannot be found or a search that matches nothing, the request costs nothing. Check that line on the page for the source you use; do not assume it for all 128.

The monthly plans on the pricing page sell points in bundles, so the planning question is how many records you pull a month. A weekly check of 50 search results across 20 queries is 1,000 points a week. A daily lookup of 200 specific products is 1,000 points a day. Those are arithmetic examples, not benchmarks.

Limits

• —Public data only. You get what a visitor sees. Anything behind a login, such as a seller's private dashboard or your own order data, is not here.
• —A snapshot, not a feed. A lookup is made at request time and does not push changes to you, so tracking price movement means scheduling repeat calls and storing results yourself. Our price monitoring guide covers that pattern.
• —Rankings move. The Amazon search source page notes that rank order can shift between requests minutes apart and that sponsored placements can appear.
• —Fields differ by marketplace. The envelope is uniform, the records are not. Read each source page before mapping it.
• —Coverage follows the site. Long-tail queries can return fewer results than you asked for.

How to decide

Before you commit to this or any other provider, check the following.

• —Is the job reading public listings? Price monitoring (the same search or product lookup on a schedule, stored as a series), competitor and assortment research (search, seller and seller-product lookups before entering a category) and agent tools (a small set of lookups so an assistant can answer a product question from live data) all fit. Managing your own listings does not.
• —Which marketplaces and regions exactly? Ask for the list, including regional sites. "Global coverage" is not an answer.
• —What does a record contain? Look at a captured response, not a feature list. Check field names and whether prices are numbers or strings.
• —Is the response shape the same across sources? A uniform envelope saves you real code. Check whether errors, partial results and costs are reported consistently.
• —How is it priced, and what is not billed? Find out whether you pay per request or per result, and what happens on a miss.
• —How fresh is a result? Find out whether lookups are made at request time or served from stored data, and what the provider says about delay.
• —Can an agent use it? If you need MCP, check whether the tools are live, not just announced.

Frequently asked questions

How many marketplaces does NeuralVerge cover?

The e-commerce category of the source catalog lists 128 sources spread across 24 marketplaces and platforms. A "source" here is one lookup, such as a product, a search, a seller or a reviews feed, so one marketplace usually has several. The catalog page for each source states exactly which sites and regions it works on.

What does an ecommerce data API call cost?

Every e-commerce source is priced in points. Lookups that resolve one record, such as a product, a seller or a category tree, cost 5 points, which the source pages state as $5 per 1,000. Search, list and review sources cost 1 point per returned result, which is $1 per 1,000. Where a source page says misses are not billed, a request that finds nothing costs nothing.

Can an AI agent call these sources through MCP?

Yes. Each source is available over REST and as an MCP tool on the hosted MCP server. The tool name is the REST function name with underscores, for example run_amazon_product_search, and the arguments match the REST body.

Is the data real time?

The source pages describe lookups as made at request time, so you are not reading a stored snapshot. That does not make the result a guarantee: a marketplace can change a price or remove a listing a minute after you read it, and search ranking can differ between two requests made close together.

Is this the same as an official marketplace API?

No. Official seller or partner interfaces from a marketplace are built around managing your own account and listings. These sources return what is publicly visible on listing, seller and category pages, in one consistent envelope. If you need to change your own listings or read your own orders, you still need the marketplace's own tools.

How do I choose between REST and MCP?

Use REST for scheduled jobs, pipelines and anything that runs without a person or an agent deciding what to fetch. Use MCP when an AI agent should pick the lookup itself, for example choosing between a product search and a seller lookup based on the question it was asked. Both reach the same sources and the same prices.