The product catalog API for AI agent commerce — search, compare, and track prices across 900,000+ merchants in the US and Southeast Asia.

BuyWhere is an agent-native product catalog API indexing 300M+ products from 900,000+ merchants across Singapore, Malaysia, Indonesia, Thailand, the Philippines, Vietnam, and the United States. It is purpose-built for AI shopping agents: BM25-ranked search, structured price comparison, deals discovery, and affiliate link tracking out of the box. The API is MCP-compatible and works with Claude Desktop, Cursor, LangChain, CrewAI, and any MCP-enabled AI client.

Get an API key at buywhere.ai/api-keys, then:

export BUYWHERE_API_KEY="bw_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
export BUYWHERE_BASE_URL="https://api.buywhere.ai"

Search products across platforms

curl -sS --get "$BUYWHERE_BASE_URL/v1/products" \
  -H "Authorization: Bearer $BUYWHERE_API_KEY" \
  --data-urlencode "q=wireless headphones" \
  --data-urlencode "limit=5"

Get a specific product by ID

curl -sS "$BUYWHERE_BASE_URL/v1/products/78234" \
  -H "Authorization: Bearer $BUYWHERE_API_KEY"

Deals feed — biggest discounts right now

curl -sS --get "$BUYWHERE_BASE_URL/v1/deals" \
  -H "Authorization: Bearer $BUYWHERE_API_KEY" \
  --data-urlencode "min_discount=20" \
  --data-urlencode "limit=10"

Build an AI agent using BuyWhere MCP tools and win API credits, featured placement, and a Built With BuyWhere badge.

Tools: search_products, get_product, compare_prices, find_deals, browse_categories, get_category_products, get_deals

Timeline: Submissions open through May 19, 2026

Enter the Challenge → | Quick Start | MCP Setup

BuyWhere is listed in the awesome-mcp-servers registry. Connect to Claude Desktop, Cursor, Windsurf, or any MCP-compatible AI client in seconds.

Install the MCP server:

pip install httpx mcp
python /path/to/buywhere-api/mcp_server.py

Claude Desktop — add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "buywhere": {
      "command": "python",
      "args": ["/path/to/buywhere-api/mcp_server.py"],
      "env": {
        "BUYWHERE_API_KEY": "your_api_key_here",
        "BUYWHERE_API_URL": "https://api.buywhere.ai"
      }
    }
  }
}

Cursor — add to Cursor settings → MCP servers using the same JSON config above.

Available MCP tools: search_products, get_product, compare_prices, get_deals, find_deals, browse_categories, get_category_products.

BuyWhere uses hybrid search (BM25 keyword + vector cosine similarity via RRF) as the default search mode. Embeddings are built with Qwen3-Embedding-4B (1024-dim) via Flow AI, replacing the retired Gemini pipeline.

• Hybrid doesn't beat keyword on the 200-query eval set → switch off
• Matching doesn't lift multi-merchant coverage → keep vectors, pause matching

• ALL embedding calls go through Flow AI only — never DeepInfra/Gemini/SiliconFlow directly
• Scope = in-stock AND price > 0 products only (~8M of 115M total)
• Only the embed worker may write to product_embeddings (write-access lock enforced)
• Nightly pg_dump of product_embeddings → R2 (30-day retention)
• model_ver = 'flow-embed-1@1024' stamped on every vector

Rate limit status is returned in response headers (X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset). On 429 Too Many Requests, use exponential backoff starting at 2 seconds.

BuyWhere is a Python/FastAPI service backed by PostgreSQL and Redis, with platform-specific scrapers deployed as ECS Fargate tasks. The scraping pipeline handles 40+ platforms concurrently using distributed Redis locks, NDJSON normalization, and BM25-ranked search via PostgreSQL FTS5.

Architecture details: SCRAPING_ARCHITECTURE.md

# Local development
docker-compose up
# API available at http://localhost:8000

Proprietary — © 2026 BuyWhere. All rights reserved.