Walmart unveils Agent Gateway, an MCP server for AI agents to shop its site, and Agent Profiles to authorize agents and track their purchases separately (Sarah Perez/TechCrunch)

Walmart's Agent Gateway is a system for managing transactions involving AI agents on Walmart's site. It is described as an MCP server, meaning it gives agents a structured way to connect with Walmart's shopping capabilities. The goal is to let software handle shopping tasks while Walmart controls how those interactions occur.

Agent Profiles add a separate identity layer for the agent. A customer can authorize an agent through that profile, and purchases made by the agent can be tracked separately from the customer's ordinary shopping activity. For example, an agent could search Walmart's catalog, act on the customer's instructions, and place an order using its authorized profile.

The provided article text does not specify every supported action, approval step, payment rule, or launch limitation. It does establish Walmart's broader direction: the company is embracing agentic commerce and building tools to manage agent transactions. That matters because shopping agents need both access to products and clear records of what they did.

An AI agent would use Walmart's Agent Gateway as its connection to the retailer's shopping system. Through that connection, it could request product information, compare possible items, and carry out shopping steps instead of merely describing what a customer should buy. The gateway is the control point between the agent and Walmart's commerce tools.

A typical example would be a customer asking an agent to find a particular household item. The agent could search Walmart's available products, select an option that fits the request, and submit an order using the customer's authorization. Agent Profiles provide the separate identity used to authorize the agent and record its purchases.

The supplied article text does not confirm the precise commands, payment flow, inventory checks, or human approval requirements. Those details may depend on Walmart's implementation. Its central mechanism is clear, however: the Agent Gateway connects the AI agent to Walmart's site, while the Agent Profile helps Walmart identify and track the agent's activity.

MCP, or Model Context Protocol, is an open standard for connecting AI applications with outside tools, services, and information. Instead of building a separate, custom integration for every agent, a service can expose capabilities through a common protocol. That gives an AI system a clearer way to discover and use available functions.

For Walmart, an MCP server can act as the bridge between an AI agent and shopping operations. An agent might use that bridge to search Walmart's product catalog, retrieve relevant details, or submit an authorized transaction. The protocol does not make the purchasing decisions itself. It provides the communication structure through which the agent can request actions from Walmart's system.

This is useful because commerce requires more than conversational answers. It requires controlled access to products, orders, and transactions. Walmart's Agent Gateway applies that connection to its site. The supplied article excerpt does not list the exact MCP tools Walmart exposes, but it identifies the gateway as an MCP server built for AI-agent shopping.

Walmart's potential reach for agent-powered shopping comes from the breadth of its retail operation. An agent connected to Walmart could potentially serve people who already shop through the company's website, while Walmart's stores and broader retail network add physical-world reach. A large catalog would also give agents many kinds of products to search and purchase.

For example, one customer might use an agent to find groceries, household supplies, or another item available through Walmart's shopping system. If the agent is authorized, the same basic connection could support transactions across the retailer's eligible assortment. The Agent Gateway is the technical access point, and Agent Profiles separate the agent's activity from ordinary purchases.

The provided article text does not state Walmart's store count, customer count, website traffic, or catalog size. Therefore, no reliable numerical estimate can be made from this excerpt. The defensible conclusion is qualitative: Walmart is applying agent commerce to a large retail ecosystem rather than to a narrow specialty shop.

An Agent Profile gives an AI agent a separate identity for Walmart shopping. The customer can authorize that profile, allowing Walmart to distinguish actions taken by the agent from actions made directly by the customer. This creates a clearer record of which purchases came through automated activity.

For example, if an agent orders several approved household items, those transactions can be associated with the Agent Profile rather than blended invisibly into the customer's regular shopping history. Walmart can then track the agent's purchases separately. Authorization also provides the basis for deciding what the agent is permitted to do, although the excerpt does not describe the available controls.

The article text does not say whether separate tracking changes the payment method, loyalty benefits, returns process, or legal responsibility. It establishes two specific effects: the agent receives authorization through its profile, and purchases made through it are tracked separately. Those records could make automated shopping easier to review, but further consequences are not stated.

Responsibility is not answered in the provided article excerpt. Walmart's launch of Agent Gateway and Agent Profiles explains how agents may be connected, authorized, and tracked, but it does not state who bears the loss when an agent selects the wrong item, exceeds instructions, or buys something without permission.

The separate Agent Profile could help investigators identify the agent involved and review its purchase history. It might also show which authorization was used. Those records can clarify the event, but recordkeeping alone does not determine liability. An agent's action could result from a customer instruction, a software error, a compromised account, or a failure in Walmart's system.

The correct answer from this source is therefore that responsibility remains unspecified. Walmart's terms, the customer's agreement with the agent provider, security practices, and applicable consumer law would be needed to decide a particular case. The article does not provide those rules, so assigning blame or financial responsibility would go beyond the available facts.

An AI agent is software designed to work toward a goal. It can interpret a request, choose steps, use connected tools, and continue until it completes the task or needs help. In shopping, that could mean finding products and submitting an order through an authorized retail connection. The agent is not limited to producing text.

A chatbot that only responds to questions might tell someone which products could fit a need. An agent could take the next steps: search a catalog, compare options, use the customer's instructions, and place a purchase through Walmart's Agent Gateway. Its ability to act depends on the tools and permissions it has been given.

Walmart's announcement illustrates this difference by focusing on transactions, not just conversation. Agent Gateway is intended to manage an AI agent's connection to Walmart's shopping system, while Agent Profiles authorize and track the agent's purchases. The supplied excerpt does not define AI agents formally, but this distinction follows from the functions it describes.