AI-powered agent that autonomously orders pizza from Domino's Taiwan (dominos.com.tw) based on natural language prompts in Chinese or English. Uses StraitsX MCP virtual card for payment — card details are retrieved at checkout time via the view_virtual_card tool.
git clone https://github.com/Xfers/dominos-agent.git && cd dominos-agent
npm install
npm run setup # installs Chromium for Playwright
cp .env.example .env # edit with your credentials
node order.mjs '2人份、想吃雞肉口味、預算800元、外帶'# Natural language prompt (card filled but NOT submitted by default)
node order.mjs '4人份、無海鮮、NT$1500以下、要可樂+副餐'
# English works too
node order.mjs 'Order for 4, seafood, budget $1500, with coke and sides, pickup'
# Submit payment for real (money moves!)
node order.mjs '2人份、想吃雞肉口味、預算800元、外帶' --submit
# Pre-configured test scenarios
npm run test:pickup # 2 people, chicken, NT$800, pickup
npm run test:delivery # 4 people, seafood, NT$1000-1500, delivery, noon
npm run test:budget # 1 person, anything, under NT$500
npm run test:vegetarian # 3 people, vegetarian, NT$1200, with drink┌─────────────────────────────────────────────────────────────────┐
│ Natural Language Prompt │
│ "4人份、海鮮口味、介於NT1000到NT1500、可樂+副餐、中午外送" │
└──────────────────────────────┬──────────────────────────────────┘
▼
┌──────────────────────────────────────────────────────────────────┐
│ 1. Parse Constraints │
│ people=4, budget=[1000,1500], wantSeafood, wantCola, │
│ wantSide, isDelivery, isLunch(12:00) │
└──────────────────────────────┬───────────────────────────────────┘
▼
┌──────────────────────────────────────────────────────────────────┐
│ 2. Store Selection (delivery or pickup, open or closed/預約) │
│ → Auto-selects time closest to desired hour │
│ → Handles store-closed with fallback to pickup │
│ → Delivery address from .env USER_ADDRESS │
└──────────────────────────────┬───────────────────────────────────┘
▼
┌──────────────────────────────────────────────────────────────────┐
│ 2.5. Promotion Scan (當期主打, 精選套餐, 自由任你配, 專屬優惠) │
│ → Checks /menu/deals for active BOGO and combo deals │
│ → Prefers 買大送大 (NT$550起) for 2+ pizza orders │
│ → Falls back to regular menu if no usable promo │
└──────────────────────────────┬───────────────────────────────────┘
▼
┌──────────────────────────────────────────────────────────────────┐
│ 3. Dynamic Menu Scan │
│ → Reads ALL pizza items + prices from live page │
│ → Filters by constraints (seafood/chicken/veg/budget) │
│ → Randomized selection (different each run!) │
└──────────────────────────────┬───────────────────────────────────┘
▼
┌──────────────────────────────────────────────────────────────────┐
│ 4. Add to Cart + Checkout │
│ → Adds pizzas, drinks (from menu), side dish │
│ → Fills contact info, selects credit card payment │
└──────────────────────────────┬───────────────────────────────────┘
▼
┌──────────────────────────────────────────────────────────────────┐
│ 5. StraitsX MCP → Payment Gateway │
│ → Calls view_virtual_card via JSON-RPC │
│ → Opens card in visible tab (demo/audit) │
│ → Extracts card number, expiry, CVV from iframe │
│ → Fills Paydollar payment form │
│ → Submits only with --submit flag │
│ → Waits up to 90s for 3DS auto-approval │
└──────────────────────────────────────────────────────────────────┘
| Constraint | Chinese | English | Example |
|---|---|---|---|
| People | N人, N個人, 一~十 | N people, for N, feed N | 4人份, feed 6 |
| Budget max | N元/以下/NT$ | under/below/max $N | NT$1500以下, under $800 |
| Budget range | 介於X到Y | between X to Y | 介於NT1000到NT1500 |
| Seafood (want) | 海鮮/蝦/蟹 | seafood/shrimp/crab | 想吃海鮮 |
| No seafood | 無海鮮/過敏 | no seafood/allergy | 無海鮮（過敏） |
| Chicken | 雞肉 | chicken | 想吃雞肉口味 |
| Vegetarian | 素食/蔬菜 | vegetarian/veggie | 素食 |
| Cola/drink | 可樂/飲料 | cola/coke/drink | 要可樂, with drinks |
| Side dish | 副餐/副食 | side/sides/appetizer | 要副餐, with sides |
| Delivery | 外送/送到 | deliver | 外送到信義區 |
| Pickup | 外帶 | pickup | 外帶 |
| Lunch | 中午/午/12 | lunch/noon | 明天中午12點 |
| Dinner | 晚/傍晚/18-20 | dinner/evening | 傍晚6點 |
| Multi-pizza | 再加一個/兩個披薩 | another/2 pizzas | 再加一個雞肉的 |
| Multi-drink | 兩瓶/三杯 | 2 bottles/3 cans | 帶兩瓶可樂 |
The agent calls view_virtual_card via JSON-RPC at payment time:
POST https://card.straitsx.ai/mcp
{
"jsonrpc": "2.0",
"method": "tools/call",
"params": {
"name": "view_virtual_card",
"arguments": {
"passphrase": "<from .env>",
"card_opaque_id": "<from .env>",
"settlement_tx": "<from .env>"
}
}
}
Response contains an iframe_url with card details. The agent:
- Opens the card in a visible browser tab (for demo recording/audit)
- Extracts card number, expiry (MM/YY), and CVV from the iframe
- Fills the Paydollar payment gateway form
- With
--submit: waits up to 90s for 3DS auto-approval by StraitsX
Full MCP response is logged to console for debugging.
| Variable | Required | Description |
|---|---|---|
CARD_MCP_URL |
Yes | StraitsX MCP endpoint |
CARD_PASSPHRASE |
Yes | Passphrase for view_virtual_card |
CARD_OPAQUE_ID |
Yes | Card opaque ID (identifies which card to view) |
CARD_SETTLEMENT_TX |
Yes | Settlement transaction hash (proves funding) |
USER_NAME |
Yes | Name for checkout (Chinese) |
USER_PHONE |
Yes | Phone for checkout (09xxxxxxxx) |
USER_EMAIL |
Yes | Email for checkout |
USER_ADDRESS |
No | Delivery/pickup address (default: 信義區) |
INVOICE_CARRIER |
No | Taiwan e-invoice carrier ID |
All tests run end-to-end: store selection → menu scan → cart → checkout → payment gateway → card filled. Items are randomized each run (different pizzas, drinks, sides) while satisfying all constraints.
| # | Prompt | Items Ordered | Total | Verdict |
|---|---|---|---|---|
| 1 | 今天晚餐想吃披薩，我們家三個人，不要太貴大概一千塊左右就好，幫我訂外帶 | 奶油白醬烤雞, 小農田園鮮蔬 | NT$965 | PASS — 2 pizzas for 3人, ≤$1000, pickup ✓ |
| 2 | 老婆說想吃有蝦子的披薩，再加一個雞肉的，順便幫我帶兩瓶可樂，預算抓兩千以內，外帶 | 日式章魚燒, 奶油白醬烤雞, 1.25L可樂 ×2 | NT$1,105 | PASS — seafood+chicken pizza ✓, 2 drinks ✓, ≤$2000 ✓ |
| 3 | 4人份、其中一個要有海鮮口味、金額要介於NT1000到NT1500、必須買可樂跟隨便一個副餐、明天中午12點外送到預先指定的地址 | 蟹肉鮮蝦沙拉, 超級美國, 香烤雞條, 1.25L可樂 | NT$1,070 | PASS — seafood ✓, side ✓, drink ✓, $1000-1500 ✓, delivery ✓ |
| 4 | 5人份、無海鮮、想吃雞肉、金額要介於NT1500到NT2500、必須買可樂跟副餐、傍晚6點外帶 | 奶油白醬烤雞, BBQ雞肉, BBQ雞肉(火山), 玉米濃湯2杯, 1.25L可樂 | NT$1,500 | PASS — 3 chicken pizzas ✓, no seafood ✓, side ✓, drink ✓, ≥$1500 ✓ |
| 5 | 4人份、無海鮮（過敏）、預算NT$1500以下、要可樂+副餐、外帶 | 在地食鮮總匯, 奶油白醬烤雞, 香烤雞條, 1.25L可樂 | NT$1,065 | PASS — no seafood ✓, side ✓, drink ✓, ≤$1500 ✓ |
| 11 | 極致干貝海鮮大手拍+香草奶油烤雞小帕瑪滋心+鱈魚星星+青花椒雞塊+香烤薯球+可樂+優惠券+週五七點外送 | 極致干貝海鮮(大/手拍), 香草奶油烤雞(小/帕瑪滋心), 鱈魚星星12塊, 青花椒雞塊, 香烤薯球, 1.25L可樂 | NT$1,134 | PASS — exact order mode ✓, coupon auto-applied ✓, Fri 7PM delivery ✓ |
| # | Prompt | Items Ordered | Total | Verdict |
|---|---|---|---|---|
| 6 | Order pizza for 4 people, seafood, budget under NT$1500, with coke and sides, pickup | 蟹肉鮮蝦沙拉, 超級美國, 香烤雞條, 1.25L可樂 | NT$1,190 | PASS — seafood ✓, side ✓, drink ✓, ≤$1500 ✓ |
| 7 | Feed 2, chicken flavor, $800 max, pickup | 奶油白醬烤雞 | NT$630 | PASS — chicken ✓, ≤$800 ✓ |
| 8 | Pizza night! 6 people, no seafood allergy, under $2000, with drinks and sides | 超級墨西哥, 肉魔王四喜, 極致蒜香壽喜牛, 香烤雞條, 1.25L可樂 | NT$1,465 | PASS — 3 pizzas for 6人 ✓, no seafood ✓, side ✓, drink ✓, ≤$2000 ✓ |
| 9 | Dinner for 5, between NT1500 to NT2500, no shrimp, want chicken, with drinks | 極致蒜香壽喜牛, BBQ雞肉, BBQ雞肉(火山), 1.25L可樂 | NT$1,180 | PARTIAL — chicken ✓, no seafood ✓, drink ✓, ≤$2500 ✓, but actual total below $1500 min (see note) |
| 10 | 3 people, vegetarian, no seafood, under 1200, with drinks | 小農田園鮮蔬, 白醬彩蔬, 1.25L可樂 | NT$680 | PASS — vegetarian ✓, no seafood ✓, drink ✓, ≤$1200 ✓ |
The agent correctly plans 3 pizzas with estimated total NT$1,590 (above budgetMin $1,500). However, Domino's live BOGO promotions sometimes discount more aggressively than our 50% estimate (e.g., buy-2-get-1-free deals), resulting in an actual total of NT$1,180. This is a known limitation: budgetMin is enforced at planning time based on menu prices, but actual checkout totals depend on live promotions that vary by day/time.
| Test | People | Budget | Flavor | Drink | Side | Time | Mode | Result |
|---|---|---|---|---|---|---|---|---|
| 1 | 3 | ≤1000 | any | - | - | dinner | pickup | ✓ |
| 2 | 2 | ≤2000 | seafood+chicken | ×2 | - | - | pickup | ✓ |
| 3 | 4 | 1000-1500 | seafood | ×1 | ✓ | noon | delivery | ✓ |
| 4 | 5 | 1500-2500 | chicken, no seafood | ×1 | ✓ | dinner | pickup | ✓ |
| 5 | 4 | ≤1500 | no seafood | ×1 | ✓ | - | pickup | ✓ |
| 6 | 4 | ≤1500 | seafood | ×1 | ✓ | - | pickup | ✓ |
| 7 | 2 | ≤800 | chicken | - | - | - | pickup | ✓ |
| 8 | 6 | ≤2000 | no seafood | ×1 | ✓ | - | pickup | ✓ |
| 9 | 5 | 1500-2500 | chicken, no seafood | ×1 | - | dinner | pickup | |
| 10 | 3 | ≤1200 | vegetarian | ×1 | - | - | pickup | ✓ |
| 11 | 2 | - | exact (seafood+chicken) | ×1 | ×3 | Fri 7PM | delivery | ✓ |
- Store closed → automatic time selection (預約) matching desired hour
- Delivery fails → fallback to pickup
- Budget minimum not met → 3-strategy upgrade: swap cheapest pizza → add extra pizzas → allow duplicates (BOGO-aware, +5% buffer)
- Dessert pizzas excluded for 3+ people (not a real meal)
- Randomized selection — pizza, drink, and side are all randomized each run
- React Native Web overlay workaround for Pressable elements (Domino's uses RNW)
- Pizza click retry — if modal doesn't open on first click, retries with coordinate-based click
- Domino's BOGO discount (~50% off 2nd+ pizza) factored into budget estimation
- noSeafood enforced on both pizzas AND side dishes
- Chinese numeral parsing — 兩千, 一千五, 三個人, etc.
- Multi-drink support — navigates to drinks menu for 2nd+ drinks (upsell cards disappear after first add)
- Explicit pizza count — "再加一個雞肉的" correctly orders 2 pizzas
- 3DS auto-approval — waits up to 90s for StraitsX 3DS verification
- MCP response logging — full JSON-RPC response printed for debugging
- Promotion scanning — checks 當期主打, 精選套餐, 自由任你配, 專屬優惠 for BOGO/combo deals before ordering
- Delivery address from .env — uses
USER_ADDRESSinstead of hardcoded value; prompt can override - Exact order mode — specify exact pizza names, sizes, crusts, and side items; agent matches precisely instead of randomizing
- Auto coupon application — applies best available BOGO voucher (買大送大/買大送小) from pizza menu page; adapts to live promotions
dominos-agent/
├── order.mjs # Main orchestrator (~1100 lines)
│ ├── parsePrompt() # NLP constraint extraction (CN + EN)
│ ├── scanPromos() # Promotion/coupon scanning (deals page)
│ ├── pickPizzas() # Budget-aware, constraint-filtered, randomized
│ ├── addPizza() # Navigate menu, click with retry, select size, add
│ ├── Store selection # Delivery/pickup, open/closed handling
│ ├── Checkout # Contact fill, side/drink from suggestions
│ └── Payment # MCP card retrieval, Paydollar form fill, 3DS wait
├── adapters/
│ ├── interface.mjs # Adapter contract (extensibility)
│ ├── dominos-tw.mjs # Domino's Taiwan (reference)
│ └── pizzahut-tw.mjs # Pizza Hut Taiwan (stub)
├── .env.example # Credential template
├── Dockerfile # Containerized execution
└── package.json # Pinned deps (playwright 1.52.0)
Key Design Decisions:
- Dynamic menu scanning — reads live prices from DOM, not hardcoded
- BOGO-aware budgeting — 2nd+ pizzas estimated at 50% off for accurate budget checks
- Randomization within constraints — shuffles by price tier, picks randomly from valid options
- MCP at payment time — card details retrieved only when needed (secure, auditable)
- No payment submission by default —
--submitflag required for real transactions - Retry with fallback — pizza clicks retry up to 3× with coordinate-based fallback
# 1. Clone and setup
git clone https://github.com/Xfers/dominos-agent.git
cd dominos-agent
npm install && npm run setup
# 2. Configure credentials
cp .env.example .env
# Edit .env with your StraitsX card credentials and contact info
# 3. Run any test (browser opens, watch the automation)
node order.mjs '4人份、無海鮮、NT$1500以下、要可樂+副餐、外帶'
# 4. Observe:
# - Store selection (pickup/delivery)
# - Menu scanning (all pizzas + prices logged)
# - Decision making (constraint satisfaction logged)
# - Cart building (items added one by one)
# - Checkout (contact filled, payment selected)
# - MCP card retrieval (visible in 2nd tab)
# - Payment form filled (NOT submitted without --submit)The browser stays open 10 seconds after completion for visual inspection (30s with --submit). Screenshots are saved to ./screenshots/.
MIT