Build an AI Shopping Assistant with BuyWhere MCP in 10 Minutes (2026)

A copy-paste walkthrough that turns Claude Desktop (or any MCP-compatible client) into a price-comparison shopping agent over 134M+ products across 75K+ merchants — under 10 minutes, no credit card.

What you'll build: A working MCP client that answers questions like "Find the cheapest 65-inch TV under $800 in stock at a US merchant with free shipping" by combining Claude's reasoning with live BuyWhere catalog data.

What you need: Node 18+ and a BuyWhere API key (free, 1,000 calls/month, no credit card). Sign up at POST https://api.buywhere.ai/v1/auth/register or via the dashboard.

Why MCP for shopping assistants

Most LLM shopping demos break at the same seam: the model knows products from training data, but it has no way to ask the live catalog. MCP (Model Context Protocol) closes that seam — the model calls structured tools, the tools return real prices, the model reasons over real data.

BuyWhere exposes 6 MCP tools, all backed by a single REST catalog. You don't pick a transport — you get the same answer whether the agent calls MCP or REST.

Step 1 — Get a free API key (60 seconds)

curl -X POST https://api.buywhere.ai/v1/auth/register \
  -H "Content-Type: application/json" \
  -d '{"email":"[email protected]","tier":"free"}'

Response:

{
  "api_key": "bw_live_•••",
  "tier": "free",
  "monthly_limit": 1000,
  "reset_at": "2026-07-01T00:00:00Z"
}

Save the key — it's the only auth header you need. Free tier gives 1,000 calls/month, no credit card.

Step 2 — Install the MCP server (90 seconds)

The official npm package handles the SSE transport, tool registration, and auth header. One install, one config entry.

npm install -g @buywhere/mcp-server

Or use npx without a global install:

npx -y @buywhere/mcp-server

Verify the package exists:

npm view @buywhere/mcp-server
# name: @buywhere/mcp-server
# version: 1.x.x

The package spins up an MCP server that proxies to https://api.buywhere.ai/mcp over streamable HTTP — your client just needs the URL and bearer token.

Step 3 — Wire it into Claude Desktop (3 minutes)

Open ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows). Add:

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

Restart Claude Desktop. You should see a 🔌 icon and a "buywhere" server with 6 tools: search_products, get_product, get_price, compare_prices, get_affiliate_link, get_catalog.

Tip: Any MCP-compatible client works the same way — Cursor, Continue.dev, Cline, Zed, custom agents. The MCP server URL is https://api.buywhere.ai/mcp and the auth header is Authorization: Bearer bw_live_•••.

Step 4 — Test with a real shopping question (2 minutes)

Open a new Claude conversation and ask:

"I'm looking for a 65-inch TV under $800 from a US merchant with free shipping and at least 4-star reviews. Compare the top 3 options and recommend one."

Claude will:

1. Call search_products with filters: q=65 inch TV, country=US, max_price=800, shipping=free, min_rating=4.
2. Call get_price on the top 3 results to confirm live availability.
3. Call compare_prices to get the side-by-side breakdown.
4. Return a recommendation with affiliate links.

The whole flow is < 5 seconds. Every price is live — the model never falls back on its training memory.

Step 5 — Call the same data from custom code (3 minutes)

MCP is the easiest path, but sometimes you need REST (serverless, custom UIs, non-MCP clients). Same data, same auth:

// search.js — find AirPods Pro under $250 in SG, sorted by price asc
const r = await fetch(
  'https://api.buywhere.ai/v1/products/search?' + new URLSearchParams({
    q: 'airpods pro',
    country: 'SG',
    currency: 'SGD',
    max_price: 250,
    in_stock: 'true',
    sort: 'price_asc',
    limit: 10
  }),
  { headers: { Authorization: `Bearer ${process.env.BUYWHERE_API_KEY}` } }
);
const { data } = await r.json();
console.log(`${data.length} results, cheapest: SGD ${data[0].price}`);

// compare.js — side-by-side for 3 product IDs
const ids = [123, 456, 789].join(',');
const r = await fetch(
  `https://api.buywhere.ai/v1/products/compare?ids=${ids}`,
  { headers: { Authorization: `Bearer ${process.env.BUYWHERE_API_KEY}` } }
);
const { data } = await r.json();
for (const p of data) {
  console.log(`${p.title} — SGD ${p.price_range.min}–${p.price_range.max} (${p.merchant_count} merchants)`);
}

The REST endpoints and MCP tools return the same payload shape, so a quick swap between transports won't break your client.

What "live" actually means

Pulled at 20:22 UTC on 2026-06-22 from GET /v1/catalog/stats:

The catalog is fed by continuous merchant ingestion — the same maglev pipeline that powers the MCP tools. When a merchant updates a price or goes out of stock, the next search_products call reflects it within minutes.

Where to go next

• MCP guide with code samples — https://api.buywhere.ai/docs/guides/mcp
• OpenAPI spec — https://api.buywhere.ai/docs
• GitHub examples (Claude Desktop, Cursor, serverless, custom agents) — https://github.com/BuyWhere/buywhere
• Pricing — https://buywhere.ai/pricing (Free 1K/mo · Starter $9/50K · Pro $49/500K · Enterprise custom)
• SLA — 99.5% uptime on production

Honest tradeoffs

BuyWhere is not a fit for every shopping agent:

• Pre-orders and unlisted SKUs aren't covered until merchants publish them.
• Out-of-stock propagation is minutes, not seconds — for sub-minute tick data, build a direct merchant integration.
• Per-merchant historical data beyond what merchants publish is out of scope.
• Shipping fee accuracy is best-effort — final cart total may differ at checkout.

For 95% of consumer shopping intents, the catalog is the right backbone. For the other 5%, the docs list direct merchant API options.

Built and verified by the BuyWhere AEO/Content team. Questions: [email protected]. Canonical URL: https://buywhere.ai/blog/buywhere-mcp-integration-walkthrough-2026.