# UCP vs ACP: Which Agentic Commerce Protocol Should You Implement First?

*Last updated: March 2026*

Two open protocols are defining how AI agents buy products on behalf of consumers. The **Universal Commerce Protocol (UCP)**, created by Google and Shopify, and the **Agentic Commerce Protocol (ACP)**, created by OpenAI and Stripe, both aim to standardize agentic commerce -- but they take fundamentally different architectural approaches, carry different fee structures, and serve different agent ecosystems.

If you are a technical leader or ecommerce manager trying to decide where to invest engineering time first, this guide provides the definitive side-by-side comparison you need to make that decision.

---

## Side-by-Side Protocol Comparison

| Dimension | UCP | ACP |
|-----------|-----|-----|
| **Created by** | Google + Shopify | OpenAI + Stripe |
| **Announced** | January 2026 (NRF conference) | September 2025 (ChatGPT Instant Checkout) |
| **Architecture** | Decentralized, merchant-hosted | Centralized, platform-controlled |
| **Discovery** | `/.well-known/ucp` manifest on merchant domain | Product feed submission to OpenAI catalog |
| **Transports** | REST, MCP, A2A, AP2, embedded | REST, MCP |
| **Payment methods** | Google Pay, Shop Pay, Stripe, Adyen, PayPal (planned), custom processors | Stripe exclusively (SharedPaymentToken) |
| **Platform fee** | None | 4% per transaction |
| **Payment processing** | Varies by processor (~2.9% + $0.30 for Stripe) | 2.9% + $0.30 (Stripe) |
| **Ecosystem partners** | 20+ (Etsy, Wayfair, Target, Walmart, Best Buy, Visa, Mastercard, Adyen) | OpenAI, Stripe, select merchants |
| **Agent surfaces** | Google AI Mode, Gemini, any open agent | ChatGPT Instant Checkout |
| **Geographic availability** | Global | US only (expanding in 2026) |
| **License** | Apache 2.0 | Apache 2.0 |
| **Implementation time** | 8-16 hours (existing backend) | 2-4 hours + review wait |
| **Approval process** | None (self-serve) | Application and review required |
| **Spec location** | [ucp.dev](https://ucp.dev/) | [agenticcommerce.dev](https://www.agenticcommerce.dev/) |

---

## UCP Deep Dive: The Decentralized Standard

### How It Works

UCP follows the same discovery pattern as `robots.txt` and `sitemap.xml`. Merchants host a JSON manifest at `/.well-known/ucp` on their own domain. Any AI agent -- Google Gemini, Anthropic Claude, a custom shopping bot -- can discover this manifest through a standard HTTP GET request without registering with a central authority.

The protocol uses a three-tier architecture inspired by TCP/IP:

1. **Shopping Service Layer** -- Core transaction primitives: checkout sessions, line items, totals, status
2. **Capabilities Layer** -- Functional domains (Checkout, Orders, Catalog) with independent date-based versioning
3. **Extensions Layer** -- Domain-specific schemas composed on top (loyalty programs, fulfillment options, AP2 mandates)

Checkout follows a state machine with three states: `incomplete` (missing information), `requires_escalation` (needs buyer input via redirect), and `ready_for_complete` (agent can finalize programmatically). This gives merchants granular control over when human involvement is required.

### Strengths

- **No platform fee.** Merchants pay only their payment processor's standard rates.
- **Multi-transport support.** A single UCP profile can expose services through REST, MCP (Anthropic's Model Context Protocol), A2A (Google's Agent-to-Agent protocol), and AP2 (Agent Payments Protocol) simultaneously.
- **Payment flexibility.** Dynamic payment handler negotiation allows merchants to offer Google Pay, Shop Pay, Stripe, Adyen, or any supported processor -- and agents select based on cart context, location, and amount.
- **Decentralized discovery.** No gatekeeper controls visibility. Any agent can discover any UCP-enabled merchant.
- **Global availability.** No geographic restrictions on participation.
- **Extensibility.** Reverse-domain namespace convention (`com.loyaltyprovider.points`) lets any vendor add capabilities without permission.

### Weaknesses

- **Higher implementation effort.** Merchants must build and host their own API endpoints (3 core checkout endpoints minimum).
- **Waitlist access.** As of March 2026, Google AI Mode integration requires joining a waitlist, though the protocol itself is open.
- **Security responsibility.** Merchants manage their own Ed25519/ES256 signing keys, CORS configuration, and PCIv4 compliance for embedded checkout.

### Best For

Merchants who want maximum reach across agent ecosystems, need payment processor flexibility, operate at scale where the 4% ACP fee becomes prohibitive, or sell internationally.

---

## ACP Deep Dive: The Integrated Platform

### How It Works

ACP uses a hub-and-spoke model. Merchants submit product feeds to OpenAI's centralized catalog and implement four REST endpoints: Create Checkout, Update Checkout, Complete Checkout, and Cancel Checkout. Payment flows through Stripe's SharedPaymentToken (SPT) -- a single-use, time-bound, amount-restricted token that the agent provisions and the merchant redeems through their existing Stripe integration.

ChatGPT serves as the primary agent surface. When a user expresses purchase intent in a ChatGPT conversation, the platform queries its centralized product index, surfaces matching products, and handles the checkout flow. OpenAI controls product ranking and visibility within the platform.

### Strengths

- **Fastest time to market.** Technical implementation takes 2-4 hours if you already have Stripe. No endpoint hosting, no manifest configuration.
- **Access to ChatGPT's 900M+ weekly users.** The single largest AI agent surface as of March 2026.
- **Stripe handles payment complexity.** PCI compliance, fraud detection, and payment processing are managed by Stripe's existing infrastructure.
- **Evolving rapidly.** Four spec versions shipped between September 2025 and January 2026, adding capability negotiation, payment handler frameworks, and an extensions system.

### Weaknesses

- **4% platform fee on every transaction.** This is in addition to Stripe's standard processing fees.
- **Stripe lock-in.** No alternative payment processors. Merchants using Adyen, dLocal, or regional processors must maintain a parallel Stripe integration.
- **Centralized control.** OpenAI controls discovery, ranking, and merchant approval. Delisting risk exists.
- **US-only availability.** International expansion is planned but not yet available.
- **Application required.** Merchants must apply at chatgpt.com/merchants and pass a review process with unpredictable timelines.

### Best For

Merchants who prioritize speed to market, already use Stripe as their primary processor, want immediate access to ChatGPT's user base, and operate primarily in the US market.

---

## Fee Comparison: Real Numbers

Fees are the single largest differentiator at scale. Here is what each protocol costs on concrete transaction volumes.

### Per-Transaction Breakdown ($100 Sale)

| Fee Component | UCP | ACP |
|---------------|-----|-----|
| Platform fee | $0.00 | $4.00 (4%) |
| Payment processing (Stripe rates) | $3.20 (2.9% + $0.30) | $3.20 (2.9% + $0.30) |
| **Total cost** | **$3.20** | **$7.20** |
| **Effective rate** | **3.2%** | **7.2%** |

### Monthly Impact at Scale ($1M/Month GMV)

| Metric | UCP | ACP | Difference |
|--------|-----|-----|------------|
| Platform fees | $0 | $40,000 | $40,000/mo |
| Processing fees | $32,000 | $32,000 | $0 |
| **Total monthly cost** | **$32,000** | **$72,000** | **$40,000/mo** |
| **Annual cost** | **$384,000** | **$864,000** | **$480,000/yr** |

At $1M monthly GMV, the ACP platform fee alone costs $480,000 more per year than UCP. For merchants operating on typical ecommerce margins of 5-15%, this difference can represent the entirety of the profit on agent-driven sales.

Note: UCP merchants can reduce processing costs further by using processors with lower rates than Stripe's standard 2.9% + $0.30, such as negotiated enterprise rates with Adyen or regional processors. ACP merchants cannot optimize this dimension.

---

## Implementation Complexity Comparison

### UCP Implementation Timeline

| Step | Estimated Time |
|------|---------------|
| Publish `/.well-known/ucp` manifest | 2-4 hours |
| Implement 3 core REST endpoints (checkout sessions) | 4-8 hours |
| Configure Ed25519/ES256 signing keys | 1-2 hours |
| Deploy and validate | 2-3 hours |
| **Total (with existing commerce backend)** | **8-16 hours** |
| **Total (from scratch)** | **2-5 days** |

Shopify merchants can skip most of this by installing the native Universal Commerce Agent app, which enables UCP in under 48 hours with no custom code.

Validation is available via the official CLI tool: `npx @ucptools/validator validate -r yourdomain.com`

### ACP Implementation Timeline

| Step | Estimated Time |
|------|---------------|
| Apply at chatgpt.com/merchants | 30 minutes |
| Submit product feed | 1-2 hours |
| Connect Stripe account | 30 minutes |
| Implement 4 checkout API endpoints | 2-4 hours |
| **Total (technical work)** | **2-4 hours** |
| **Review and approval** | **Days to weeks** |

Shopify and Etsy merchants have pre-existing eligibility and can bypass the application process entirely.

### Infrastructure Requirements (Both Protocols)

| Requirement | Target |
|-------------|--------|
| Product discovery response time | Under 200ms |
| Checkout completion response time | Under 500ms |
| Error rate | Under 1% |
| Uptime | 99.9%+ |
| Inventory sync | Real-time (sub-second) |

Agents skip slow merchants. Performance is not optional -- it directly impacts whether agents recommend your products. According to early UCP benchmark data published by UCP Hub, merchants with response times above 500ms see measurably lower agent discovery rates.

---

## The Dual-Protocol Strategy: 40% More Agentic Traffic

UCP and ACP are not competing standards that force an either-or decision. They serve different agent ecosystems with minimal overlap, and dual-protocol merchants are already seeing measurable results.

Etsy, the most prominent dual-protocol merchant, reports up to **40% more agentic traffic** compared to single-protocol stores (source: UCP vs ACP comparison data from DEV Community and industry analysis, early 2026). The reason is straightforward: UCP captures traffic from Google AI Mode, Gemini, Anthropic Claude, and any open-standard agent. ACP captures traffic from ChatGPT. These audiences are largely additive.

### Shared Infrastructure

Both protocols require the same underlying commerce infrastructure:

- Product Catalog API (structured product data with real-time inventory)
- Checkout Session API (cart management, pricing, tax, shipping)
- Order Management API (status tracking, fulfillment updates)
- Payment Processing (tokenized payment handling)

Once you build these capabilities for one protocol, adding the second is incremental. The recommended total for both protocols is 10-20 hours of engineering time for a merchant with an existing commerce backend.

---

## Decision Framework

### Choose UCP First If...

- You sell internationally or plan to expand beyond the US
- You process more than $100K/month in GMV where the 4% ACP fee becomes material
- You want payment processor flexibility (not locked into Stripe)
- You want exposure to Google AI Mode, Gemini, Claude, and the broader open-agent ecosystem
- You are on Shopify and can use the native Universal Commerce Agent app
- You need custom extensions (loyalty programs, marketplace features, domain-specific capabilities)
- You value decentralized discovery without gatekeeper approval

### Choose ACP First If...

- You already use Stripe as your primary payment processor
- Your primary market is the United States
- You want the fastest possible time to first agentic sale
- ChatGPT is the dominant AI tool among your target customers
- You are on Shopify or Etsy with pre-existing eligibility (bypassing the review process)
- Your GMV is below $50K/month where the 4% fee is acceptable relative to the speed advantage
- You prioritize simplicity over flexibility

### For Most Merchants: UCP First, Then ACP

The recommended approach for the majority of merchants is to implement UCP first, then add ACP. The rationale:

1. UCP provides broader agent coverage (Google, Anthropic, open agents) with no platform fee
2. Building UCP endpoints creates the commerce API infrastructure that ACP also requires
3. Adding ACP after UCP is operational takes only 2-4 additional hours
4. Total investment for both protocols: 10-20 hours

---

## How UCP and ACP Work Together

These protocols are complementary layers in a broader agentic commerce stack, not competing alternatives. The full emerging stack includes:

| Protocol | Purpose | Creator |
|----------|---------|---------|
| **UCP** | Full commerce lifecycle (discovery through post-purchase) | Google + Shopify |
| **ACP** | Agent-initiated checkout and payment delegation | OpenAI + Stripe |
| **AP2** | Trust, authorization, and cryptographic mandates | Google |
| **MCP** | Agent-to-tool communication | Anthropic |
| **A2A** | Agent-to-agent coordination | Google |

A dual-protocol merchant does not duplicate work. The product catalog, inventory system, checkout logic, and order management are shared. Each protocol is a distribution channel -- a different door through which AI agents enter your store.

The market signal is clear: AI-driven orders grew 15x from January 2025 to January 2026 according to Shopify data. Agentic traffic converts at 19.8% on average versus 2.8% for organic search -- nearly 9x higher, according to early 2026 UCP benchmark data from UCP Hub. Cart abandonment drops from approximately 70% for human shoppers to under 5% for agent-driven sessions. These numbers make the case that supporting both protocols is not a question of if, but when.

---

## Frequently Asked Questions

**Do I need to choose one protocol exclusively?**

No. UCP and ACP are complementary, not mutually exclusive. Both are open standards under Apache 2.0. Etsy supports both simultaneously and reports 40% more agentic traffic as a result. The shared infrastructure means the second protocol is always easier than the first.

**What is the total cost difference at $1M monthly GMV?**

UCP costs approximately $384,000/year in payment processing fees (assuming Stripe rates). ACP costs approximately $864,000/year -- $480,000 more due to the 4% platform fee. Merchants using processors with rates lower than Stripe's standard pricing would see an even larger gap.

**Can I use ACP without Stripe?**

Not currently. ACP's payment flow is built on Stripe's SharedPaymentToken (SPT). PCI DSS Level 1 merchants can alternatively use proprietary vaults, but Stripe remains the only supported payment provider. UCP supports multiple processors including Google Pay, Adyen, and custom integrations.

**Is ACP available outside the United States?**

As of March 2026, ACP is US-only with international expansion planned. UCP has no geographic restrictions and is available globally.

**How long does the ACP review process take?**

Timelines vary. Shopify and Etsy merchants have pre-existing eligibility that bypasses the standard review. For other merchants, the application at chatgpt.com/merchants can take days to weeks. UCP has no approval process -- merchants self-serve by publishing a manifest on their own domain.

**What conversion rates should I expect from agentic commerce?**

Early 2026 benchmark data from UCP Hub shows average agentic conversion rates of 19.8%, with high-intent queries reaching 29.5%. This compares to 2.8% for organic search and 1.2% for social media referral. The Sierra Roast case study documented a reduction in customer acquisition cost from $40 to $4 after implementing UCP.

**Do I need to rebuild my storefront for agentic commerce?**

No. Your online store (traditional web for human browsers) and your agentic storefront (backend API for AI agents) are separate surfaces. Neither protocol requires theme modifications or frontend changes. UCP requires hosting a JSON manifest and API endpoints. ACP requires submitting a product feed and connecting Stripe.

---

## Recommendation

For technical leaders evaluating agentic commerce in March 2026, the evidence points to a clear strategy: implement UCP first for its broader ecosystem reach, zero platform fee, payment flexibility, and global availability. Then add ACP to capture ChatGPT's user base with minimal incremental effort.

The exception is merchants who are US-only, already on Stripe, and prioritize speed above all else. For them, ACP's 2-4 hour implementation (assuming pre-existing eligibility) can deliver the first agentic sale faster.

Regardless of which protocol comes first, the goal is dual-protocol support. The 40% traffic increase that dual-protocol merchants report is too significant to leave on the table. The underlying commerce infrastructure is shared, the incremental engineering cost is modest, and the agentic commerce market is growing at a pace where early movers will have a structural advantage.

Agentic traffic converts at nearly 9x the rate of traditional search. Cart abandonment drops from 70% to under 5%. Customer acquisition costs can fall by an order of magnitude. The question is no longer whether to implement these protocols, but how quickly you can get both live.

---

*Sources: [ucp.dev](https://ucp.dev/), [agenticcommerce.dev](https://www.agenticcommerce.dev/), [Shopify Engineering](https://shopify.engineering/ucp), [Google Developers Blog](https://developers.googleblog.com/under-the-hood-universal-commerce-protocol-ucp/), [OpenAI Commerce Documentation](https://developers.openai.com/commerce/), [Stripe ACP Documentation](https://docs.stripe.com/agentic-commerce/protocol), [UCP Hub Benchmarks](https://ucphub.ai/agentic-commerce-conversion-rate-ucp/), [DEV Community Technical Comparison](https://dev.to/ucptools/ucp-vs-acp-in-2026-a-technical-comparison-of-ai-commerce-protocols-50j7)*