Baci, ChatGPT and MCP: From Product Discovery to Checkout

AI assistants can interpret structured product feeds and use connected tools to query a store, but those are separate integration paths. Baci documents ACP and UCP commerce foundations; that does not establish a live ChatGPT connection for every merchant. Nigerian sellers should first verify catalog accuracy, permissions, payment eligibility and delivery coverage.

Decision summary

For merchants exploring ChatGPT MCP shopping agents for African stores, the useful question is what an assistant can reliably learn and do. Can it identify the correct variant? Can it obtain current availability? Can it establish the delivery charge before asking the buyer to approve a purchase? Each question needs its own evidence.

Baci’s mission is to build the future of commerce in Africa by making practical commerce infrastructure more accessible, starting with Nigeria. Its documented combination of catalog, order, payment and fulfillment workflows gives that ambition a concrete foundation. Agent interfaces extend that direction, but implementation is separate from verified merchant activation or measured commercial impact.

This guide reviews official documentation as of 6 November 2026 and the supplied Baci capability register reviewed on 28 October 2026. It does not report a completed live checkout test. The recommendation is to improve the information and operating processes behind your store now, then validate any agent connection against the exact account and workflow you intend to use.

What MCP does, and how it differs from crawling

Model Context Protocol, or MCP, standardizes how an AI application exchanges information with external systems. In the protocol’s architecture, the application connects through an MCP client to a server that exposes capabilities. Those capabilities can include tools for requesting information or performing defined actions.

In a hypothetical commerce integration, a developer might expose a product-search tool and a separate availability lookup. An assistant could request products matching a buyer’s requirements, receive structured results, and use those results in its answer. These are illustrative tool functions, not verified Baci tool names.

A crawler visiting a product page follows a different route. It reads accessible web content rather than necessarily calling a connected merchant tool. Likewise, a product feed supplies organized catalog records through an agreed delivery process. A merchant should ask which route an integration actually uses instead of treating every AI interaction as MCP.

OpenAI’s MCP documentation describes listing a connected server’s tools and calling selected tools. That requires a configured connection. Publishing a storefront or a protocol profile does not, by itself, connect that store to every ChatGPT conversation.

Conceptual illustration of structured information exchange, rather than a depiction of a live Baci or ChatGPT interface. No text or logos.

Feature comparison: pages, feeds, MCP and commerce protocols

These mechanisms solve different parts of the same problem. The following decision table compares their roles, rather than assigning unsupported platform scores.

Route	What it contributes	What merchants must verify	Baci evidence boundary
Public product pages	Readable product information for shoppers and accessible web discovery.	Correct descriptions, variant links and consistent storefront information.	A public store does not prove inclusion in an assistant’s recommendations.
Structured product feed	Catalog records with identifiable attributes such as price and availability.	Destination schema, required fields, refresh process and acceptance.	Product-feed integration surfaces do not prove an accepted OpenAI feed.
MCP connection	Defined tools and information exposed to a connected AI application.	Actual server, available tools, authentication and permitted actions.	The reviewed register does not establish a universal merchant-facing Baci MCP connection.
ACP integration	A documented commerce integration path for structured discovery and transaction flows.	Partner access where required and the exact enabled integration.	Baci’s implementation evidence does not establish OpenAI partner approval.
UCP profile and services	Machine-readable declaration and negotiation of commerce capabilities.	Compatible versions, declared services and authorization requirements.	Implemented profiles do not establish complete current-spec conformance.
Checkout execution	A controlled transition from selected products to an order.	Final totals, payment configuration, buyer authorization and recorded outcome.	No end-to-end production checkout was verified for this article.

UCP’s official documentation describes discovery through a machine-readable profile at /.well-known/ucp. Compatible participants determine which capabilities they share. Its documentation also distinguishes services from their transport bindings, including REST and MCP. Consequently, UCP support and MCP support should not be treated as interchangeable claims.

The supplied Baci register records implementation of ACP and UCP discovery profiles, structured catalog interfaces, UCP carts, checkout-session operations and authorized order reads. Discovery is merchant-scoped and capability-gated. The public Baci features overview provides the appropriate starting point for discussing those foundations with the platform team.

How discovery should lead to a reliable recommendation

OpenAI’s official commerce onboarding guide starts with a structured product feed. It identifies titles, descriptions, images, prices and availability as useful catalog information, and currently limits feed onboarding to approved partners. The guide recommends refreshing a full snapshot at least daily. Those requirements describe OpenAI’s integration process; they do not prove Baci has completed it for a particular store.

For a merchant, the first practical task is making records unambiguous. Consider an illustrative Nigerian clothing store selling one design in several sizes. A general description saying “available in all sizes” is not enough when only particular variants remain. A recommendation should refer to the purchasable variant and its applicable offer.

Separate product identity from promotional language. Maintain clear titles, stable identifiers, accurate variant attributes and consistent currency information. If a photograph depicts several items but the offer contains only one, describe exactly what is included. These are editorial recommendations for reducing ambiguity, not a claim about mandatory fields in every Baci interface.

Freshness is a second decision. A feed snapshot and a later stock lookup can disagree when something sells between them. Before enabling purchase actions, ask the integration owner when availability is rechecked and what happens if the selected variant becomes unavailable. An assistant should not be expected to infer missing inventory information from persuasive copy.

The third decision is recommendation scope. A catalog connection provides information; it does not guarantee that an assistant will recommend an item, prefer one merchant or generate demand. No topic-specific positive demand measurements were supplied for this assignment, so there is no basis here for projecting traffic, rankings or sales.

Checkout and payments: eligibility comes before automation

A product match is only the beginning of a transaction. For a Nigerian merchant, the operational questions include whether the configured payment provider accepts the transaction and whether the destination can be served. A conversational interface cannot resolve an unsupported payment configuration simply by describing it confidently.

Baci’s reviewed evidence covers merchant-configured Paystack, Korapay, Credit Direct and pay-on-delivery options, subject to country, currency, eligibility and provider setup. It also covers shipping zones, rates, pickup and supported shipment workflows. These are conditional capabilities, not promises that every option is enabled for every merchant.

Before an agent-assisted pilot, request a demonstration using your intended provider, currency and delivery route. Confirm how the final delivery charge is obtained, where the customer sees the payable total and which action authorizes payment. Verify the resulting order and payment state rather than stopping at a successful-looking response.

Baci’s agent checkout implementation requires readiness checks, signed authorization and enabled payment configuration. OpenAI separately states that Instant Checkout in ChatGPT is available to approved partners. Neither an ACP endpoint nor a configured local gateway establishes admission to that program.

Keep discovery access separate from sensitive actions. A tool that returns public product information needs a different permission boundary from one that reads a customer order or initiates checkout. OpenAI’s MCP guidance supports tool filtering and configurable approval requirements; the appropriate policy must be established for the actual integration. This article does not verify those settings for a deployed Baci merchant connection.

Conceptual illustration: product discovery, payment authorization and fulfillment require separate checks. No readable text, logos or UI chrome.

Pricing and cost model

The Baci pricing page, checked on 6 November 2026, lists Free at ₦0, Pro at ₦5,000 per month and Premium at ₦15,000 per month. Free lists one store and a maximum of 10 products. Pro lists unlimited products and custom-domain support; Premium lists multiple staff accounts.

Those published labels do not establish the subscription entitlement for ACP, UCP or a custom MCP connection. The supplied register warns that some runtime tier names differ from the public pricing vocabulary. Obtain an explicit entitlement confirmation instead of mapping plan names yourself.

Budget separately for the store subscription, any required integration work, domain expenses, applicable payment-provider charges, delivery and ongoing maintenance. Pro’s published “No Transaction Fees” wording should not be read as evidence that an external processor charges nothing. No provider-fee schedule or complete agent-integration price was verified here.

A useful quotation should identify who maintains the feed or server, handles authentication changes and investigates failed requests. Compare the cost of maintaining the workflow with the business task it solves. A catalog lookup pilot and a payment-capable agent integration have different scopes, so their quotations should describe those differences.

Best fit: prepare the catalog, then pilot a bounded workflow

This approach best fits merchants and developers who can name a specific information problem: confusing variants, inconsistent product records or uncertainty about delivery eligibility. It is less useful to commission broad “AI shopping” work before deciding which question the system must answer correctly.

Start with a small set of representative products. Include a product with variants, an unavailable item and an offer with a delivery restriction. Ask a connected assistant factual questions and compare its answers with the underlying records. These are proposed acceptance checks; they have not been performed for this article.

Next, test failure handling. What does the integration return when stock is unknown, a provider is unavailable or a customer lacks access to an order? The desired result is a clear limit and a usable next step. A plausible answer that conceals uncertainty is a poor basis for taking payment.

Assign ownership before expanding the pilot. Someone must maintain product data, review exceptions and decide when a workflow is ready for customers. Baci’s broader commerce platform is the context for this work: discovery needs dependable operations behind it. Public marketing claims alone should not replace account-specific verification.

Verdict

Prepare for agent-mediated commerce by improving the catalog and validating a narrow, authorized workflow. Structured feeds organize product information; MCP connections expose callable tools; ACP and UCP address commerce-specific interactions. Baci has documented and code-evidenced foundations in this direction, while live merchant activation, external approval and successful checkout remain separate verification steps.

The practical contribution to Baci’s Africa-commerce mission is connecting discoverable products to workable local operations. Accurate variants, eligible payments and serviceable delivery routes are concrete priorities for Nigerian merchants. Their value does not depend on predicting when conversational shopping will become widespread, and this review establishes no measured adoption or revenue impact.

FAQ

Does every Baci store automatically appear in ChatGPT shopping?

No automatic inclusion is established. Baci’s features page explicitly separates protocol support from admission to ChatGPT or Google shopping. OpenAI’s commerce onboarding guide, checked on 6 November 2026, says product-feed onboarding is available to approved partners.

Is MCP the same as ACP or UCP?

No. MCP standardizes exchanges between AI applications and connected servers. ACP and UCP address commerce interactions. UCP can describe services using different transport bindings, including MCP, but support for one protocol does not establish support for every capability of another.

Can an AI agent take payments through any Baci gateway?

That is not established. Gateway eligibility, merchant configuration, checkout readiness and authorization apply. A supported payment-settings workflow does not prove that a specific agent can successfully execute a payment. Verify the exact provider and end-to-end flow before relying on it.

Which Baci plan includes an MCP shopping connection?

The reviewed public pricing does not establish a plan entitlement for an MCP shopping connection. Confirm the enabled workflow, subscription requirements and any integration charges with Baci. Do not infer access from the Free, Pro or Premium labels alone.

Sources

• Baci Features — conditional commerce capabilities and ACP/UCP interfaces
• Baci Pricing — public plan prices checked 6 November 2026
• OpenAI Agentic Commerce — Get Started
• OpenAI Agentic Commerce — Key Concepts
• OpenAI API — MCP Servers
• Model Context Protocol — Architecture Overview
• Universal Commerce Protocol — Core Concepts