
Image: Walmart
Why it mattersAn agent that signs in through a merchant-run MCP endpoint can shop without screen-scraping, which is the first retailer-side pattern that works at the primitive level agent builders need.
Retailers have spent the past month blocking AI shopping agents one by one. Walmart is going the other way. The retailer said it will let AI agents shop on its website starting next month, through a new "Walmart Agent Gateway" that uses the Model Context Protocol, with a per-agent profile, spending caps and approval rules built in.
Walmart's announcement sets out three pieces that work together. Agent Profiles give each AI agent its own profile inside a customer's Walmart account, so the agent's purchases appear separately from the person's own. The Walmart Agent Gateway is a beta MCP endpoint that agents call instead of navigating a web browser. Terms for AI Agents are the rules an agent must follow to use the gateway at all. Together they are the first retailer-run framework that treats an agent as a first-class user of a shopping system, rather than as a browser instance to tolerate or block.
What the gateway actually gives agents
Walmart says agents currently "may shop on Walmart's site through a web browser, following steps designed for people. This can be slow and error-prone, and agents may get stuck along the way." The gateway replaces that with an MCP connection, which Walmart says is "a faster, more reliable way to shop."
The gateway also lets an agent look up a customer's usual items without reading the full purchase history. Walmart's example: "if a customer tends to buy sourdough bread or a particular brand of milk, Walmart will help the agent find those usual items for the customer without sharing the customer's broader shopping history with the agent." The beta is for companies and developers.
What the customer sees
Approval is required for every agent purchase by default. A customer can also set a spending limit per agent, such as $50, $100 or $200, and the agent can transact up to that cap without further approval. Agent purchases are listed separately from the customer's own, and the customer can delete an Agent Profile at any time.
Walmart's terms spell out what agents cannot do: take actions beyond what the shopper intends, change a payment method, or access information beyond what the task needs. Walmart says it will block agents that break these rules. The service covers Sam's Club too.
The news contrasts with Amazon, which sued Perplexity over its shopping agent and recently blocked Meta's Muse. Walmart is choosing to publish a protocol instead, and the choice of MCP matters: it is the same primitive agent builders at OpenAI, Anthropic and others are already wiring their agents against.
For a team building an agent that transacts, this changes the stack. A merchant-run MCP endpoint with explicit per-agent identity, spending caps and audit trails is what has been missing from every "agents can shop" demo so far, and the Walmart launch is the first time a retailer this size has published one.
Source
Making It Possible for Agents to Shop on Behalf of Customers, Walmart. Reported by Emma Roth at The Verge.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.

