5 Things You Can Build with UCP Today

UCP isn't just a specification to read — it's infrastructure you can build on today. The Registry API and thousands of live merchant endpoints mean you can go from idea to working prototype in hours. Here are five projects worth building, with enough detail to get you started.

#1. AI Shopping Assistant

The most obvious and most valuable application. Build an AI chatbot that takes natural language shopping queries and returns real products from real merchants.

How it works:

1. User says: "Find me wireless earbuds under $100 with noise cancellation"
2. Your agent queries the Registry API for electronics merchants with dev.ucp.shopping.catalog capability
3. Agent searches each merchant's catalog for matching products
4. Results are ranked by price, reviews, or relevance and presented to the user

What you need:

• Registry API access for merchant discovery
• An LLM for natural language understanding
• UCP client library (Python or JavaScript) for merchant queries

Our full tutorial on building an AI shopping agent walks through this end to end with working code.

Why it's worth building: AI shopping assistants are the primary use case driving UCP adoption. Every major AI platform is investing here, and the merchant base is large enough to be useful.

Real-world data: In 560+ tracked agent sessions, shopping assistants powered by 16 different AI models successfully reached checkout 47% of the time. The average session takes just 36 seconds and 6.7 turns — fast enough for real-time conversational commerce. Models like Gemini 3 Flash and Llama 3.3 70B achieve 71% checkout success rates, proving this is a solved problem, not a research project.

#2. Price Monitoring Tool

Track prices across UCP-enabled merchants and alert users when prices drop. Unlike traditional price scrapers that break constantly, UCP gives you a stable, structured API.

How it works:

1. User adds a product (or product category) to their watchlist
2. Your tool queries relevant merchants via UCP on a schedule
3. When a price drops below the user's threshold, send a notification
4. Optionally, use the agent to auto-purchase at the target price

What you need:

• A list of merchant endpoints (from the Registry)
• A scheduler (cron job, queue worker) for periodic checks
• dev.ucp.shopping.catalog capabilities from merchants
• Storage for price history

Why it's worth building: Price monitoring is a billion-dollar market (Honey, CamelCamelCamel, etc.), and UCP makes it dramatically easier to build. No scraping, no reverse-engineering APIs, no dealing with anti-bot measures.

Real-world data: UCP product listing endpoints respond in ~980ms on average and product detail endpoints in ~270ms — fast enough for frequent price checks. With 47+ unique stores already accessible via UCP and error rates under 2% for product listing operations, the reliability is production-ready. Across tracked sessions, agents processed 24 million tokens of product data — the raw material for price intelligence. The total tracked GMV exceeds $108,000 across our sample.

#3. Product Comparison Engine

Build a tool that compares products across merchants on price, specs, availability, and shipping. Think Wirecutter meets real-time data.

How it works:

1. User searches for a product category (e.g., "standing desks")
2. Your engine queries multiple merchants via UCP
3. Results are normalized and compared across dimensions
4. User gets a structured comparison with links to purchase

What you need:

• Registry API for merchant discovery
• Product data normalization logic
• A frontend for displaying comparisons

Why it's worth building: Cross-merchant comparison has always been limited by data access. UCP removes that barrier.

Real-world data: Agent sessions show that when agents compare across merchants, they average 3-5 stores per query in routed sessions. The catalog search operation runs at ~460ms per merchant, meaning a 5-store comparison completes in under 2.5 seconds when parallelized. Cart values in comparison scenarios average $165, , suggesting users are making meaningful purchases — not just window shopping. With the payment handler ecosystem maturing, you can even enable one-click purchase from the comparison view.

#4. Inventory Availability Checker

Build a tool that checks whether a specific product is in stock across all UCP merchants that carry it. Useful for high-demand drops, limited editions, or hard-to-find items.

How it works:

1. User provides a product name or SKU
2. Tool queries the Registry for merchants in the relevant category
3. For each merchant, check product availability via dev.ucp.shopping.catalog
4. Return a real-time availability map

What you need:

• Registry API for merchant discovery
• dev.ucp.shopping.catalog capability from merchants (most support it — 93% adoption rate)
• Caching layer for frequently-checked products

Why it's worth building: This is particularly valuable for DTC brands that sell out quickly. Consumers want to know where products are available right now, and UCP makes that a single API call per merchant.

#5. UCP Compliance Dashboard

Build a tool for merchants to monitor and improve their UCP implementation. Show them their score, what's missing, and how to improve.

How it works:

1. Merchant enters their domain
2. Tool fetches their /.well-known/ucp.json manifest
3. Validates the manifest against the UCP spec
4. Tests each declared capability for correctness and speed
5. Generates a report with scores and improvement suggestions

What you need:

• UCP spec knowledge for validation rules
• HTTP client for endpoint testing
• Benchmark logic (the Registry's approach is a good reference)

Why it's worth building: Merchants want to know how their UCP implementation compares and what to fix.

Real-world data: Error rate analysis across 2,100+ tool calls reveals clear patterns. Product listing has the lowest error rate at 2%, while checkout operations see 4-30% error rates. The most common errors are type mismatches (passing integers where strings are expected) and missing identity tokens. A diagnostic tool that catches these issues before they frustrate AI agents would be immediately valuable to the 47+ stores already in the ecosystem — and the thousands more joining. The Registry provides benchmark scores, but a detailed diagnostic tool with actionable recommendations is a different, complementary product. If you're a Shopify merchant, this is especially useful for verifying your setup.

#Getting Started

All five projects build on the same foundation:

1. Registry API — documented here — for merchant discovery
2. UCP manifests — fetched directly from merchant domains
3. Merchant capabilities — dev.ucp.shopping.catalog, dev.ucp.shopping.cart, dev.ucp.shopping.checkout

The UCP SDKs (Python, JavaScript, Go) handle the protocol details so you can focus on your application logic.

Pick a project, build a prototype, and join the ecosystem. The infrastructure is ready — the interesting part is what you build on top of it.

#Which AI Model to Use

If you're building an agent, model choice matters. Session data across 16 models shows wide performance variance:

Model	Checkout Success Rate	Avg Tokens	Best For
Gemini 3 Flash	71%	36,500	Cost-efficient, fast
Llama 3.3 70B	71%	61,000	Open source, self-hosted
DeepSeek V3	70%	28,400	Most token-efficient
Claude Sonnet 4.5	66%	67,800	Best reasoning, complex flows
Claude Opus 4.6	63%	41,800	Balanced performance
Grok 4	57%	34,600	Good mid-range option

The takeaway: more expensive doesn't mean more successful. DeepSeek V3 achieves 70% success with the lowest token usage. Gemini 3 Flash matches the top success rate at a fraction of the cost of frontier models. Choose based on your latency, cost, and hosting requirements.

---

Start with the developer documentation, browse live merchants to test against, or submit your own tool to the Registry.