UCP Shopping Agent

AI Shopping Agent that discovers UCP merchants, compares prices across vendors, optimizes multi-merchant orders, and orchestrates autonomous purchases -- like what powers Google AI Mode in Search and Gemini.

Buyer-side UCP implementation. While the UCP Merchant Server shows how merchants serve the protocol (like Shopify), this project shows how AI agents consume it to shop across multiple stores.

---

What This Project Demonstrates

Concept	Implementation
UCP Discovery	Fetch /.well-known/ucp from multiple merchants in parallel
Multi-Merchant Search	Fan-out product search across all discovered merchants
Price Comparison	Build comparison matrix with price, shipping, and ratings
Split-Order Optimization	Buy items from cheapest vendors to minimize total cost
LangGraph Workflow	9-node state graph: plan -> discover -> search -> compare -> optimize -> confirm -> checkout -> complete
Human-in-the-Loop	Confirmation step before any purchase is executed
SSE Streaming	Real-time shopping progress events
MCP Tool Surface	All operations exposed as MCP tools
3 Mock Merchants	Built-in TechZone, HomeGoods, MegaMart for instant demo

---

Quick Start

Docker Compose (Recommended)

git clone https://github.com/samuelvinay91/ucp-shopping-agent.git
cd ucp-shopping-agent

cp .env.example .env
# Edit .env with your API keys (optional - works without LLM for basic flows)

docker compose up --build

The API will be available at http://localhost:8020. Docs at http://localhost:8020/docs.

Local Development

git clone https://github.com/samuelvinay91/ucp-shopping-agent.git
cd ucp-shopping-agent

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

python -m ucp_shopping.main

---

Shopping Workflow

User: "Find me the best deal on a mechanical keyboard and a USB-C hub"

    [PLAN] Parse intent -> extract items + constraints
      |
    [DISCOVER] Fetch /.well-known/ucp from 3 merchants
      |
    [SEARCH] Fan-out parallel search across all merchants
      |
    [COMPARE] Build price/shipping comparison matrix
      |                          +---------------------------+
      |                          | Keyboard    | USB-C Hub   |
      |                          |-------------|-------------|
      |                          | TechZone    | $79 + $5.99 |
      |                          | HomeGoods   | $89 + FREE  |
      |                          | MegaMart    | $69 + $8.99 |
      |                          +---------------------------+
      |
    [OPTIMIZE] Split-order: keyboard from MegaMart ($69), hub from HomeGoods ($34)
      |
    [PRESENT] Show comparison + recommendation via SSE
      |
    [CONFIRM] Wait for human approval
      |
    [CHECKOUT] Create sessions at MegaMart + HomeGoods simultaneously
      |
    [COMPLETE] Aggregate orders -> unified tracking

---

API Reference

Shopping (Main Flow)

# Start a shopping session
curl -X POST http://localhost:8020/api/v1/shop \
  -H "Content-Type: application/json" \
  -d '{"query": "Find me a mechanical keyboard under $100"}'

# Check shopping progress
curl http://localhost:8020/api/v1/shop/{session_id}

# Stream real-time progress (SSE)
curl -N http://localhost:8020/api/v1/shop/{session_id}/stream

# Confirm purchase
curl -X POST http://localhost:8020/api/v1/shop/{session_id}/confirm

# Cancel shopping
curl -X POST http://localhost:8020/api/v1/shop/{session_id}/cancel

Price Comparison

# Compare a product across all merchants
curl -X POST http://localhost:8020/api/v1/compare \
  -H "Content-Type: application/json" \
  -d '{"query": "gaming mouse"}'

# Optimize a multi-item order
curl -X POST http://localhost:8020/api/v1/optimize \
  -H "Content-Type: application/json" \
  -d '{"items": ["mechanical keyboard", "USB-C hub", "monitor stand"]}'

Merchant Discovery

# List known merchants
curl http://localhost:8020/api/v1/merchants

# Discover a new merchant
curl -X POST http://localhost:8020/api/v1/merchants/discover \
  -d '{"urls": ["https://merchant.example.com"]}'

# Browse a merchant's catalog
curl http://localhost:8020/api/v1/merchants/techzone/catalog

Order Tracking

# List all orders across merchants
curl http://localhost:8020/api/v1/orders

# Get specific order
curl http://localhost:8020/api/v1/orders/{order_id}

---

Built-in Mock Merchants

Three UCP-compliant mock merchants are included for instant demo:

Merchant	Products	Specialty	Port Path
TechZone	20 electronics	Laptops, keyboards, mice	/merchants/techzone
HomeGoods	20 home/office	Desk accessories, lighting	/merchants/homegoods
MegaMart	20 general	Overlapping catalog, different prices	/merchants/megamart

Each merchant serves /.well-known/ucp, product catalog, checkout sessions, and orders.

---

SSE Events

When streaming a shopping session (GET /api/v1/shop/{id}/stream), you'll receive these events:

event: planning
data: {"message": "Parsing shopping request...", "items": ["keyboard", "hub"]}

event: merchants_discovered
data: {"count": 3, "merchants": ["TechZone", "HomeGoods", "MegaMart"]}

event: searching
data: {"merchant": "TechZone", "query": "keyboard"}

event: products_found
data: {"merchant": "TechZone", "count": 5}

event: comparison_ready
data: {"matrix": [...], "best_price": {...}}

event: optimization_ready
data: {"plan": {...}, "savings": "$12.50"}

event: awaiting_confirmation
data: {"total": "$103.99", "merchants": 2}

event: checkout_progress
data: {"merchant": "MegaMart", "status": "completed"}

event: completed
data: {"orders": [...], "total": "$103.99"}

---

Testing

pytest tests/ -v
pytest tests/ -v --cov=src/ucp_shopping
pytest tests/ -v -m "not slow and not integration"

---

Project Structure

ucp-shopping-agent/
├── src/ucp_shopping/
│   ├── main.py                    # Entry point + mount mock merchants
│   ├── api.py                     # FastAPI routes
│   ├── config.py                  # Settings
│   ├── models.py                  # All Pydantic models
│   ├── streaming.py               # SSE event stream
│   ├── orchestrator/
│   │   ├── graph.py               # LangGraph shopping workflow
│   │   ├── state.py               # Graph state schema
│   │   └── planner.py             # LLM-powered intent parser
│   ├── protocols/
│   │   ├── ucp_client.py          # UCP protocol client
│   │   ├── a2a_bridge.py          # A2A agent integration
│   │   └── mcp_surface.py         # MCP tool definitions
│   ├── agents/
│   │   ├── discovery_agent.py     # Merchant discovery
│   │   ├── search_agent.py        # Multi-merchant search
│   │   ├── comparison_agent.py    # Price comparison matrix
│   │   ├── optimizer.py           # Split-order optimization
│   │   └── checkout_agent.py      # Multi-merchant checkout
│   └── mock_merchants/
│       ├── merchant_factory.py    # Create mock UCP merchants
│       ├── merchant_app.py        # Reusable merchant mini-app
│       └── catalogs/              # Product data (JSON)
│           ├── techzone.json
│           ├── homegoods.json
│           └── megamart.json
├── tests/
├── k8s/
├── Dockerfile
├── docker-compose.yml
└── pyproject.toml

---

Key Technologies

• UCP - Universal Commerce Protocol (agent-side client)
• LangGraph - Multi-step shopping workflow
• A2A - Agent-to-Agent protocol bridge
• MCP - Model Context Protocol tool surface
• FastAPI - API framework + SSE streaming

---

License

MIT License - see LICENSE for details.