The protocol stack of agentic commerce: ACP, AP2 and MCP

Agentic commerce does not work by having an AI agent read a human web page and guess. It works through protocols — contracts that define how an agent requests a catalog, carries a payment intent, and pulls live context. A protocol gives the agent a dependable surface that does not break when a page layout changes or a rate limit triggers.

Three protocols matter today, each championed by a different model maker and each covering a distinct part of the transaction. ACP from OpenAI covers discovery and checkout, AP2 from Google covers payments, and MCP from Anthropic covers live context. Together they describe the full arc of an agent-mediated purchase, from question to paid order.

For a brand, the implication is direct: a catalog reachable on only one protocol is reachable to only part of the market. As the category creator, SkuLift designed its Agentic Commerce Platform to be protocol-plural, exposing one canonical catalog across all three so the brand is reachable wherever the buyer's agent lives.

ACP, the Agentic Commerce Protocol, is associated with OpenAI and ChatGPT. It defines how a shopping agent discovers products and completes a checkout, so a buyer inside ChatGPT can move from a question to a purchase without leaving the assistant. It is the protocol that puts a brand's catalog inside the largest consumer AI surface.

For a brand, supporting ACP means its products can be surfaced and transacted against by ChatGPT's commerce features. SkuLift maps the brand's canonical catalog onto ACP so the listings are discoverable and the checkout path is complete, and it tracks the protocol's evolving fields so the integration stays compliant. The dedicated ACP page covers the mechanics in depth.

AP2, the Agent Payments Protocol, is associated with Google and Gemini. It addresses the hardest part of autonomous commerce: letting an agent move from recommending a product to actually paying for it on the buyer's behalf, with verifiable authorization and the right guarantees. AP2 is what turns an agent's recommendation into a settled order.

SkuLift supports AP2 so that a brand reachable through Gemini and Google's agents can complete the transaction, not merely appear in the answer. Because the platform holds one canonical catalog, the same prices and policies that an agent reads also govern the payment, keeping the transaction consistent. The dedicated AP2 page goes further.

MCP, the Model Context Protocol, is associated with Anthropic and Claude and is now adopted far beyond it. MCP lets an agent call external tools and pull live context — for commerce, querying a catalog for the current price, availability and policy at the moment of the answer rather than relying on stale training data. MCP keeps an agent's answer accurate and current.

SkuLift exposes a brand's catalog as an MCP-callable source, so Claude and other MCP clients can fetch live product truth when they answer. Combined with ACP and AP2, this completes the stack: pull context over MCP, present a product, and pay via ACP or AP2. The dedicated MCP page explains the integration in detail.

Rather than asking a brand to build and maintain three separate integrations, SkuLift projects a single canonical catalog onto all three protocols. The brand integrates one source of truth; the platform keeps each protocol surface compliant as specifications evolve, and absorbs spec changes at the platform layer instead of pushing them onto the brand.

This is also a measurement boundary. The platform samples agent answers behind each protocol, so a brand can see whether it is cited in ChatGPT but missing in Gemini, or present in Claude but mispriced in ChatGPT, and fix the precise gap. Protocol plurality is how the Agentic Commerce Platform makes a brand reachable, consistent and measurable across every major agent — the category SkuLift defined and operates.