Agents
AI agents for ecommerce need real data, not vibes
1.37M
Shopify stores in the catalog
368.9M
SKUs snapshotted daily
60
MCP tools your agent can call
Source: EcomScout catalog, pulled 2026-09-04. See the numbers in use: our 2026 dropshipping niches study is built from three of these tools.
What an ecommerce AI agent can do with a catalog
These are real asks and the tools the agent picks for them. No prompt engineering — the tool descriptions do the routing.
“Who competes with gruntstyle.com and what did they change this month?”
find_peers → competitor_moves
Peer stores in the same vertical and visit band, then their price cuts, new SKUs, and app installs since a date.
“Which fitness stores in Germany are growing fastest?”
get_movers
Stores ranked by visit or revenue growth, filtered to vertical and country, with the percentages.
“Find dropship-flagged pet stores with SKUs under $20.”
screen_dropship
Flagged stores plus their cheapest SKUs — the starting list for sourcing research.
“Write an outreach list for skincare stores with an email on file.”
find_outreach → get_contacts
Stores that match the filter and actually have a contact — emails, socials, ESP tells.
The full list is in the tools reference; a static walkthrough lives on capabilities. Price-specific workflows have their own page: price intelligence.
Why we built it agent-first
Dashboards make you translate a question into filters, export a CSV, and do the thinking yourself. An agent skips the translation: it takes “who cut prices in my niche this week” straight to the rows. So we shipped the data as 60 MCP tools instead of another web app — GET-only, labeled fields, quotas on the key. The agent does the clicking.
One honest caveat: traffic and revenue are modeled estimates, not merchant-reported numbers. Every field says what it is, so the agent can cite ranges instead of inventing precision.
Common questions
What is an AI agent for ecommerce?
An assistant (Cursor, Claude, ChatGPT) with tools attached, so it can fetch real store and product data instead of guessing from training data. You ask in plain language; it decides which tool to call and reads the rows back.
Why does the agent need an MCP server?
MCP is how agents call external tools. Without one, your agent answers ecommerce questions from memory — stale and unverifiable. With Neonjelly attached it queries a live catalog of 1.37M stores and cites what it found.
Which agents work with Neonjelly?
Anything that speaks MCP: Cursor, Claude and Claude Code, ChatGPT, VS Code, Windsurf, or your own script over the REST catalog API. Install is one click from /start; auth lives in the connect URL.
What does it cost to try?
Nothing for 14 days — the trial mints on install, 100 calls a day, no account. Paid plans start at $29/mo on Stripe.
Install
Every button starts the same free trial and connects your AI in one step.
One click, 14-day trial, no account. Or test a raw call first in the playground.
Indexed by