I use Claude, ChatGPT, Grok, Gemini and Cursor, and I kept re-explaining what I’m shopping for to each one. So I built Buy-Box: you save a “buy box” (budget, must-haves, dealbreakers) once, and any agent you connect reads it before it searches or recommends.
I use it for a house search and a car search, but it works for anything: laptops, flights, furniture, gifts.
Tools: set_buy_box, get_buy_box, list_categories, score_candidates, export_buy_box, delete_buy_box
Add to Cursor (~/.cursor/mcp.json):
{
"mcpServers": {
"buy-box": { "url": "https://buybox.agentconsult.ai/mcp" }
}
}
Cursor will prompt you to sign in (OAuth, no API key or client ID to set up). Then try:
Save a buy box for a used car: under $25k, AWD, under 60k miles, no Nissans.
Later, in any connected agent:
Score these listings against my car buy box.
Details
- Remote MCP server, Streamable HTTP, nothing to install
- Your boxes are private to your account; export or delete them with a tool call
- Never collects payment details or shopping site passwords
- Listed on the official MCP Registry, cursor.directory and Glama
- Tested in Cursor, Claude, VS Code and Gemini
Free during early access. I’m looking for early users to tell me what breaks or what’s missing, so reply here with feedback.