State of the UCP Ecosystem — March 2026
We've been tracking the UCP ecosystem since the protocol's early days. Here's a snapshot of where things stand as of March 2026 — the numbers, the trends, and what they mean for the next phase of growth.
By the Numbers
| Metric | Value |
|---|---|
| Total merchants in Registry | 3,000+ |
| Verified merchants | 2,800+ |
| Platforms represented | 12+ |
| Average response time (verified) | ~180ms |
| Merchants with checkout capability | 800+ |
| Countries represented | 30+ |
| Agent sessions tracked (playground) | 560+ |
| Unique stores tested by agents | 47+ |
| Total agent tool calls observed | 2,100+ |
| Total tokens processed | 24M+ |
These numbers have roughly doubled in the past six months, driven by platform-level UCP support and growing awareness among merchants.
What the Agent Session Data Tells Us
We've been tracking real AI agent sessions interacting with UCP merchants via the UCP Playground. Across 560+ sessions using 16 different AI models, the data reveals how agents actually use UCP in practice:
- 58% of sessions reach checkout or create a cart — meaning more than half of agent interactions progress beyond browsing to transactional intent
- 47% reach full checkout — agents are successfully navigating the complete purchase flow nearly half the time
- Average session duration is 36 seconds — agents complete their entire shopping workflow (discover, search, select, checkout) in well under a minute
- Average of 6.7 turns per session — the typical agent interaction involves about seven round-trips with the merchant endpoint
- Stores expose an average of 7.5 tools — merchants are offering rich capability surfaces beyond basic search
The most telling metric: checkout success rate has been trending upward, reaching 79% on peak days in late March. This suggests both agent intelligence and merchant implementation quality are improving rapidly.
Platform Adoption
Shopify leads in absolute numbers, which isn't surprising given its market share. But the more interesting story is the breadth of platform support:
- Shopify — native UCP support, largest merchant count. See our Shopify setup guide
- WooCommerce — UCP adapter plugin available, strong adoption among WordPress merchants
- Magento/Adobe Commerce — community module with full capability support
- BigCommerce — native support rolling out
- Salesforce Commerce Cloud — enterprise UCP integration via Agentforce
- Custom platforms — growing number of custom implementations from large retailers
Capability Distribution
Not all UCP endpoints are created equal. Here's how capabilities break down across verified merchants:
| Capability | Adoption Rate |
|---|---|
dev.ucp.shopping.catalog |
95% |
dev.ucp.shopping.cart |
45% |
dev.ucp.shopping.checkout |
35% |
dev.ucp.shopping.order |
15% |
dev.ucp.shopping.returns |
8% |
Product discovery capabilities are near-universal — most merchants start here because it's the easiest to implement and provides immediate value.
Our agent session data confirms this: product listing and search account for 44% of all tool calls made by AI agents. The next most common operation is checkout creation at 10%, followed by cart updates at 5%. This distribution tells us agents spend most of their time in the discovery phase, with a focused funnel toward transactions.
Interestingly, agents call product detail tools (getting specifics on a single product) only 5.6% of the time — suggesting that structured search results are rich enough for most purchase decisions without a separate detail fetch. The most capable merchants support the full stack, and their numbers are growing as platforms add deeper UCP support.
Transport Trends
MCP is the most popular transport in the UCP ecosystem, with REST and embedded as alternatives:
- MCP: 70% of merchants
- MCP + REST or Embedded: 25% of merchants
- REST or Embedded only: 5% of merchants
MCP's dominance reflects the protocol's design — UCP was built with MCP's tool-calling paradigm as the primary interaction model, though REST provides a familiar fallback for traditional integrations. In our playground data, we've observed two dominant MCP server implementations: a storefront renderer and a WooCommerce UCP adapter — both reflecting the ecosystem's growth from Shopify and WooCommerce merchants. For more on how these protocols work together, see our deep dive on UCP, A2A, and MCP.
Benchmark Quality
Our benchmark system scores merchants across five dimensions: speed, data quality, capability coverage, manifest compliance, and agent accessibility.
The average benchmark score has improved steadily since UCP launched in January 2026:
- January 2026: Average score 62
- February 2026: Average score 68
- Early March 2026: Average score 73
- Late March 2026: Average score 76
This improvement reflects both better tooling (platforms making it easier to implement well) and growing awareness of what "good" looks like.
What "Good" Looks Like in Practice
Agent session data gives us concrete benchmarks for what high-quality UCP implementations deliver:
- Product search response time: The best implementations respond in ~100ms. The ecosystem average sits around 980ms for listing operations and 460ms for catalog search — a gap that highlights the performance advantage of optimized endpoints
- Checkout creation speed: Average of 480ms, with the fastest implementations completing in under 350ms
- Error rates: Product listing has the lowest error rate at 2%, while checkout operations see 4-30% error rates depending on implementation maturity. The most common errors are type mismatches (agents passing integers where strings are expected) and missing identity tokens
- Cart operations: Average 540ms response time with a 15% error rate — the area with the most room for improvement The gold standard collection now has more merchants than ever scoring 90+.
AI Bot Policy Trends
A significant shift since UCP launched: merchants are becoming more permissive with AI bot access. Many merchants initially blocked AI crawlers by default, but the trend has reversed quickly:
- 78% of verified merchants allow GPTBot
- 75% allow ClaudeBot
- 82% allow Google-Extended
- 70% allow all five major AI agents
The AI bot-friendly collection tracks merchants that welcome all major agents — a useful filter for developers building universal shopping assistants.
Payment Handler Integration
Payment handler support for UCP has expanded significantly:
- Stripe — most mature integration, supports Payment Links and Payment Intents
- PayPal — Express Checkout flow adapted for agents
- Visa/Mastercard — network-level verification for agent transactions
- Adyen — enterprise multi-currency support
- Splitit — installment payment support through UCP
How Different AI Models Perform with UCP
One of the most interesting findings from our agent session data is the variance across AI models. We've tracked 16 different models interacting with UCP endpoints:
| Model | Sessions | Success Rate | Avg Tokens |
|---|---|---|---|
| Gemini 3 Flash | 24 | 71% | 36,500 |
| Llama 3.3 70B | 51 | 71% | 61,000 |
| DeepSeek V3 | 37 | 70% | 28,400 |
| Claude Sonnet 4.5 | 118 | 66% | 67,800 |
| Claude Opus 4.6 | 59 | 63% | 41,800 |
| Grok 4 | 37 | 57% | 34,600 |
| Gemini 2.5 Flash | 79 | 51% | 30,800 |
| GPT-4o | 39 | 49% | 27,700 |
| Gemini 2.5 Pro | 49 | 43% | 27,700 |
Key takeaways:
- Success rates range from 25% to 71% across models — the protocol is model-agnostic, but some models navigate tool-calling workflows more effectively than others
- Token efficiency varies widely: DeepSeek V3 achieves 70% success with only 28,400 tokens on average, while Claude Sonnet uses 67,800 tokens for 66% success. More tokens doesn't always mean better outcomes
- Smaller models can compete: Gemini 3 Flash and Llama 3.3 70B both hit 71% success rates, outperforming larger models like GPT-5 (44%) and Gemini 2.5 Pro (43%)
- The ecosystem is truly multi-model: 16 different models from 7 providers are actively using UCP, confirming the protocol's vendor neutrality
This data is critical for agent developers choosing which model to power their shopping assistant. It's also valuable for merchants — your UCP endpoint needs to handle the quirks of multiple AI models, not just one.
What's Working
Platform-Level Support
The single biggest driver of adoption has been platforms adding native UCP support. When Shopify flips a switch and thousands of merchants get UCP endpoints, that's a step-change in ecosystem growth.
Developer Tooling
The UCP SDKs (Python, JavaScript, Go) have matured significantly. Building an agent that queries UCP endpoints is now a afternoon project, not a week-long integration effort. Our agent building tutorial demonstrates this.
Registry as Discovery Layer
The UCP Registry has become the default way agents discover merchants. The API handles millions of queries per day, and the benchmark system gives agents a reliable quality signal.
What Needs Improvement
Checkout Adoption
Only 35% of merchants support checkout. This limits what agents can do — full-stack agentic commerce requires the ability to complete purchases, not just discover products.
Data Quality Consistency
While average benchmark scores are improving, there's still a long tail of merchants with incomplete or outdated product data. AI agents are only as good as the data they receive.
International Coverage
The ecosystem is heavily weighted toward North America and Europe. Asian, Latin American, and African markets are underrepresented — a gap that needs closing for UCP to be truly universal.
Multi-Currency and Global Reach
Agent session data shows transactions in 4 currencies (USD, EUR, GBP, INR), confirming that UCP is being used internationally. USD dominates at 85% of cart sessions, but European and Asian transactions are growing. Average cart values vary significantly — from \ to over \ in a single session — reflecting the diversity of use cases from small DTC purchases to larger retail transactions.
Across all sessions with cart data, agents created carts averaging $165 in value with 2-3 items per cart. The total gross merchandise value tracked across our sample exceeds $108,000 — real economic activity flowing through UCP endpoints.
Looking Ahead
The next six months will be defined by:
- Checkout becoming default — platforms will enable it by default, pushing adoption above 50%
- MCP transport going mainstream — as AI platforms standardize on MCP, merchants will follow
- A2A integration — first production agent-to-agent commerce scenarios
- Enterprise adoption — Fortune 500 retailers moving from pilots to production
The foundation is solid. The protocol works, the merchants are there, and the tools are ready. 2026 is the year agentic commerce goes from early adopter to early majority.
Browse the full UCP Registry to explore the ecosystem, or check the developer API to start building.
