MCP · open · no login · stores nothing

Plug your agent into the real price of delivery.

Your agent reads the menu or the bag; UPAY3FOOD returns the 3 layers of the price, the cost per litre and per gram, the menu ranking and a recommendation with a link to the item. The same math as the extension, with a public methodology.

Connector URLhttps://upay3food.com/api/mcp

Connect

Claude (claude.ai and apps)

Settings → Connectors → Add custom connector → paste the URL.

https://upay3food.com/api/mcp

On your phone (Claude app)

Add the connector once on claude.ai and it shows up in the app. Send a screenshot of an iFood menu and ask:

Read this screenshot and use UPAY3FOOD: which item here is the best value per 100 g?

Claude Code

In the terminal:

claude mcp add --transport http upay3food https://upay3food.com/api/mcp

Cursor / VS Code / Windsurf

In the editor's mcp.json:

{
  "mcpServers": {
    "upay3food": { "url": "https://upay3food.com/api/mcp" }
  }
}

ChatGPT

With developer mode on: Settings → Connectors → Create → paste the URL.

https://upay3food.com/api/mcp

stdio-only clients

Through mcp-remote:

npx -y mcp-remote https://upay3food.com/api/mcp

There is also a pesquisar_delivery prompt that walks the agent through a search. Menu names change between app versions; what matters is adding a remote MCP server by URL (Streamable HTTP).

Tools

`price_breakdown`	You pay 3 times	food, fees and the difference to the cheapest option; the fees' share of the total
`item_cost`	Price X-ray	per 100 ml, per litre, per 100 g, per person, per 100 ml with fees, the fees in ml of the product, grams of meat
`rank_menu`	Unit prices for a menu	per 100 ml, per 100 g of meat, per 100 g of dishes sold by weight, drinks per litre; tubs and 1 kg packs apart, combo extras priced
`advise`	Menu search	“burger under R$40” → estimated total, best per 100 g of meat, link to the item
`read_menu_text`	Read a menu (text)	paste the shop page text → items, struck-through prices, servings, delivery fee and a unit-price summary
`read_bag_text`	Read a bag (text)	paste the bag text → lines, fees, a checked total and the 3 layers
`compare_bags`	Compare bags	real bags for the same purchase → ranking and the real difference to the cheapest
`parse_ifood_link`	Read an iFood link	city, shop, shop and item ids; drops tokens and tracking
`methodology`	Methodology	how each number is computed and what we never claim

Extension + agent

Open a few iFood shops with the extension. No login needed.
In the popup, ask “quero hambúrguer até R$40” (a burger under R$40) and click Copiar para o seu agente (copy for your agent).
Paste it into your agent with the UPAY3FOOD connector on. It calls advise with the menus you read and answers with the link to the item.

Search for me: "a burger under R$40".
Use the UPAY3FOOD MCP server, tool advise,
with these menus I read on iFood…

Rules for agents

Read only what the user can see. iFood shop pages open without login.
Never place an order or pay: UPAY3FOOD recommends, the user confirms in the bag.
No cookies, tokens, addresses or personal data in the calls: the tools do not need them.
Every menu result is an estimate. Coupons, membership and address change the price.

The server stores nothing and calls no platform: it only computes with what the agent sends. Technical docs · llms.txt