A tutorial shows how to give an AI agent live prices from many online shops

The wrong-price problem

Ask an AI assistant who sells a specific pair of headphones most cheaply right now, and you tend to get one of two disappointing replies. Either it guesses from what it learned in training, which may be months or years old, or it points you at a single shop. Neither helps if you are actually trying to buy, or to compare suppliers.

As dev.to reported in a tutorial, the fix is to give the agent a live data source. The example product in the guide is the Sony WH-1000XM5, and the aim is for the agent to return current listings from many shops, each with a seller, a price and a link you can check yourself.

What the server hands your agent

The tool in the tutorial is called datawebot. It is a hosted MCP server. MCP, short for Model Context Protocol, is a standard way of plugging outside tools and data into an AI assistant, so you do not have to build a custom connection for each one. According to the guide, any MCP-capable client can use it, including Claude, ChatGPT, Cursor, VS Code and Gemini CLI, and you do not write a scraper.

The agent gets three kinds of ability:

• Search. The search_products tool takes a query and a country, then returns product titles, prices, shops, ratings, review counts, images and links.
• Compare sellers. The get_product_offers tool takes one product and lists every seller with its price, a total that includes delivery charges, notes on stock or delivery, and a direct link to that seller's page.
• Pull structured data. For store pages, the agent chooses a template with list_job_blueprints, submits a web address with submit_job, and reads the records back with get_results. These jobs work on Amazon search and product pages in 10 countries, on Walmart US, and on other stores in the service's endpoint directory.

Connecting it takes one line

The quickest route is to let the agent configure itself. You paste a single instruction into Claude Code, Cursor, Codex or any agent that can read a web page, telling it to read the setup file at mcp.datawebot.com and follow the steps. The agent starts a sign-in, and you approve it from your email inbox. Nobody pastes a secret key into the chat, which is a sensible design.

If you prefer to do it by hand, the server speaks Streamable HTTP (a standard web transport) with OAuth sign-in, the same kind of login you use when an app asks to connect to your Google account. You add a short block of settings naming the server and its address. Claude Code users can alternatively install a plugin, which adds the server along with a skill, a set of instructions that teaches the agent how to use the tools well. For scripts that are not agents, a REST API with an API key and an OpenAPI spec is available.

A typical run goes in order. The agent first lists the tools available. It then searches for the product in a given country, takes the product ID from a promising result, and asks for all offers. For bulk or page-level data, it moves through the job steps instead.

What it cannot do

The tutorial is upfront about one limit. What comes back is a snapshot taken at the moment of retrieval, with links to the source. The service keeps no price history and sends no alerts. So if you want to know whether a price is a good deal over time, or to be pinged when it drops, you would need something else. Treat every figure as accurate as of the time it was fetched.

Some questions are not answered in the guide. It does not tell you how many shops the search covers beyond the named stores, and it does not explain how the service deals with sites that restrict automated access. If you plan to build a business process on this, read the terms of the shops that matter to you.

What it costs

Signing in is free. Every account gets a daily allowance of free searches, plus 200 trial credits to spend on extraction jobs. When those run out, you buy credit packs, which start at $10. There is no subscription, and jobs that fail or return nothing cost nothing.

The author also publishes the connection kit under an MIT license, a permissive open-source license, and lists the server in the official MCP Registry. That makes the plumbing easy to inspect, even though the data service itself is hosted and paid beyond the free tier.