How to Build a Cross-Border Price Comparison Agent for Southeast Asia (2026)

#ai #mcp #opensource #tutorial

Title: How to Build a Cross-Border Price Comparison Agent for Southeast Asia (2026)
Subtitle: A working MCP-powered agent that compares a single product across Shopee SG, Lazada MY, Shopee ID, Amazon SG, and 6 other marketplaces — in under 30 minutes.
Tags: ai, mcp, tutorial, opensource, agents
Canonical URL: https://buywhere.ai/blog/cross-border-price-comparison-agent-sea-2026?utm_source=devto&utm_medium=blog&utm_campaign=june30_25k&utm_content=cross_border_agent
Cover image: https://buywhere.ai/og-image.png

Why cross-border price comparison is a real problem

Singapore is the regional shopping hub. Shoppers routinely ask: should I buy this in Singapore or send it from Malaysia? Should I wait for a Lazada MY flash sale or buy now on Shopee SG? Answering that accurately requires live prices across six marketplaces in five currencies.

That is exactly what an MCP-powered agent should do. Here is a working version, end-to-end, in under 30 minutes.

What you will build

A CLI agent that, given a product query and a buyer country, returns:

The cheapest matching product across all marketplaces
The best price-per-merchant, with shipping cost estimate
A buy-now-or-wait signal (based on 30-day price history)
A canonical link for the buyer to convert

It uses the BuyWhere MCP server as the data layer and the OpenAI Agents SDK as the orchestration layer.

Step 1 — Install dependencies

mkdir sea-price-agent && cd sea-price-agent
npm init -y
npm install @buywhere/mcp-server @openai/agents

You will need a BuyWhere API key (free at buywhere.ai/api-keys) and an OpenAI API key.

Step 2 — MCP server config

Save this as mcp.json:

{
  "mcpServers": {
    "buywhere": {
      "command": "npx",
      "args": ["-y", "@buywhere/mcp-server"],
      "env": {
        "BUYWHERE_API_KEY": "${BUYWHERE_API_KEY}"
      }
    }
  }
}

Step 3 — Define the agent

Save this as agent.js:

import { Agent, Runner } from "@openai/agents";

const buywhereAgent = new Agent({
  name: "SEA Price Comparison Agent",
  instructions: `
You are a price comparison agent for Southeast Asian e-commerce.
When the user asks about a product, use the BuyWhere MCP tools to:
1. search_products(query, market="SG,MY,ID,TH,PH,VN") — get the candidate set
2. For each candidate, get_product(id) — confirm specs
3. compare_products(product_ids, include_history=true) — return best price + 30-day low
4. Reply with: cheapest option, market, price in user currency, link, and a buy-or-wait signal

Always convert prices to the user's currency (SGD, MYR, IDR, THB, PHP, VND, USD).
Always include a canonical link so the user can convert.
Never recommend a product without confirming it is in stock.
  `.trim(),
  mcpServers: ["buywhere"]
});

const runner = new Runner();
const result = await runner.run(
  buywhereAgent,
  process.argv[2] || "cheapest PlayStation 5 in Singapore right now"
);
console.log(result.finalOutput);

Step 4 — Run it

export BUYWHERE_API_KEY="bw_your_key_here"
export OPENAI_API_KEY="sk-your-key"
node agent.js "cheapest iPhone 16 Pro 256GB in Singapore vs Malaysia"

Example output

Cheapest: iPhone 16 Pro 256GB at Shopee SG — SGD 1,649 (USD 1,232).
Runner-up: Lazada MY at MYR 4,499 (≈ USD 999), but ships from overseas (5-day delivery, USD 25 shipping).
Net landed cost in Singapore: Lazada MY at ≈ USD 1,024 vs Shopee SG USD 1,232.
Recommendation: Lazada MY + shipping wins by ~USD 200, but if you need it today, Shopee SG.
Link: https://buywhere.ai/compare/iphone-16-pro-256gb?utm_source=agent

That is the working agent. The full primitives — search_products, get_product, compare_products, get_deals, list_categories — are documented at buywhere.ai/api-docs.

What this does NOT do (yet)

Real-time shipping: shipping costs are estimates. Production version would call a logistics MCP.
Cross-currency checkout: the agent recommends the buy; payment still requires going to the merchant page.
Inventory over time: the 30-day low is current; historical backfill requires the deals history tool.

These are next steps for the production version. The skeleton above is the smallest working MCP commerce agent for SEA.

Why this matters for builder economics

A working SEA cross-border price agent is a real product, not a demo:

70M Singapore, Malaysian, and Indonesian online shoppers compare prices across borders every week
The affiliate economics work: 4-8% commission on electronics, 2-3% on fashion, recurring on subscriptions
The MCP primitive (compare_products returning structured comparison data) is genuinely useful for any shopping agent

If you ship this in production, you have a real commerce agent for one of the highest-volume shopping regions in the world.