Walmart Opens Its Doors to AI Shopping Agents: MCP-Based Agent Gateway and Agent Profiles With Spending Limits, the Opposite of Amazon
Walmart announced the MCP-based Agent Gateway, Agent Profiles with spending limits, and terms for AI agents. How it differs from Amazon's blocking of outside agents, and what ecommerce merchants should decide about agent access, identification, and returns.
Key Takeaways
- On October 21, 2026, Walmart announced three building blocks for agent-led shopping: the MCP-based Agent Gateway, Agent Profiles with spending limits and purchase approvals, and a set of terms for AI agents
- In contrast to Amazon, which blocks outside agents, Walmart defines identification, permissions, and even returns for mistaken purchases up front, bringing agent-placed orders in as a legitimate sales channel
- Ecommerce merchants should decide now whether to block agents or offer an official entry point, how to identify agent orders, and how to handle returns when an agent buys the wrong thing
Walmart Builds the Entry Point, Permissions, and Rules for Agents at Once
As Amazon blocks AI agents from shopping on its site, Walmart is taking the opposite approach, introducing new technology that lets AI agents make purchases on customers' behalf, with controls over spending and transaction approvals.
techcrunch.comOn October 21, 2026 (U.S. time), Walmart President and CEO John Furner took the stage at the Masters of Scale Summit in San Francisco and announced a new framework that lets AI agents shop on customers' behalf. According to Walmart's official announcement, it covers both Walmart and its warehouse club, Sam's Club.
The announcement has three parts: Agent Profiles, which give each agent its own space in the customer's account; the Walmart Agent Gateway, a dedicated connection point for agents; and Terms for AI Agents, the rules agents must follow.
The timing matters. As TechCrunch points out, Walmart's biggest rival, Amazon, blocked Meta's AI agent Muse from shopping on its site just a month earlier. Walmart is one of Muse's retail partners and already offers integrations with ChatGPT and Gemini. This announcement formalizes agent-led shopping as a legitimate sales channel.
Agent Profiles: Separating "Who Bought This" Like an Authorized Card User
Agent Profiles is the most practical piece of the announcement. According to Walmart, agents today sign in using the customer's own account. As a result, orders placed by the customer and by the agent appear together in the purchase history, with no way to tell them apart.
With Agent Profiles, each agent gets a dedicated profile within the customer's account, and customers can view the agent's purchases separately from their own. They can choose how purchases are approved: either approve every purchase, or let the agent buy within a spending limit they set. The official announcement gives $50, $100, and $200 as example limits.
Walmart compares this to adding an authorized user to a credit card. The difference is that the agent never receives the card information itself, and the customer can turn it off at any time.
Setup takes three steps. The first time a customer shops with a given agent through the Agent Gateway, Walmart automatically creates a profile for that agent. The customer then signs in to their Walmart account, authorizes the profile, and picks a spending limit. Profiles can be deleted at any time. According to TechCrunch, the feature launches within the next month and does not require a paid Walmart+ membership.
Walmart was also explicit about pricing. Saying it prices "products, not people," the company stated that the prices an AI agent sees are the same prices a human sees on Walmart.com, with applicable offers and fees shown before purchase. The statement preempts any suspicion that customers shopping through agents might be charged differently.
On the other hand, Digital Commerce 360 reports that Walmart has not said which agents will support the profiles at launch. Walmart also says it may require additional approval even within a spending limit, but the conditions for that have not been disclosed.
Agent Gateway: From Browser Automation to MCP
The Agent Gateway is a dedicated entry point for agents connecting to Walmart. It uses MCP (Model Context Protocol), an open standard developed by Anthropic that lets AI applications connect to external tools and data.
Why does Walmart need a dedicated entry point? In Walmart's words, agents today shop through a web browser, following steps designed for people. This is slow and error-prone, and agents can get stuck along the way.
This is not a theoretical problem. A TechCrunch article from October 6 reported multiple complaints on social media that Muse could not complete purchases on Walmart's site. A Walmart spokesperson said the blocks were not intentional. According to TechCrunch, the culprit was a button that verifies the visitor is human; if that check is interrupted, it fails and the agent gets kicked out. In other words, a partner agent was being turned away by Walmart's own bot defenses. The Agent Gateway is a direct answer to that mismatch.
Daniel Danker, Walmart's executive vice president of AI acceleration, product and design, said the gateway supports discovery, add to cart, purchasing, and order tracking.
Another feature is the ability to buy a customer's "usuals" without handing the purchase history to the agent. If a customer regularly buys a particular brand of milk, Walmart helps the agent find it without sharing the customer's broader shopping history. It is a design that welcomes agents without giving up control of the data.
The gateway will begin as a beta for companies and developers. The start date, participants, and whether agent operators will pay any fees have not been disclosed.
The Terms Decide Who Bears the Cost of a Mistaken Purchase
The third pillar, the terms for agents, best captures the character of this announcement. The seven items in the official announcement fall into four areas.
On responsibility, Walmart states that its regular returns process applies, including when an agent does not act as intended. On permissions, agent purchases require customer approval by default, and agents must not take actions beyond what the shopper intends. Agents do not receive payment card information and cannot add or change payment methods.
On information, agents can access only the information needed for the task and allowed by the customer. Information an agent receives is handled under the agent provider's privacy policy. On enforcement, agents that do not follow the terms will be blocked from Walmart's site.
The returns clause carries the most weight. In a September piece, TechCrunch noted that one reason Amazon is cautious is that "if Muse makes a bad order, Amazon is going to be the one to clean it up," dealing with both the angry customer and the angry vendor. Walmart has declared up front that it will absorb that same risk through its returns process.
Why Amazon Keeps Agents Out
It is worth checking Amazon's side against primary reporting. According to GeekWire, starting on the night of September 20 (U.S. time), people trying to shop on Amazon through Muse saw the message: "Continued access by an unauthorized AI agent violates Amazon's Conditions of Use, to which our customers have agreed."
Amazon makes three arguments: it was not told that Muse would access its store, Muse does not identify itself when it browses, and it appears to capture and store customer credentials. Meta, for its part, says Muse has no visibility into people's passwords or payment methods.
Litigation is part of the backdrop. Amazon won a preliminary injunction against Perplexity over its Comet browser in March, but on August 4 the Ninth Circuit Court of Appeals vacated it, ruling that under federal anti-hacking law it is the user who is accessing Amazon's computers. That is why the warning shown to Muse users cites Amazon's Conditions of Use rather than hacking. GeekWire also notes that Amazon generated more than $68 billion in ad revenue in 2025, a business that depends on people viewing its product pages.
That said, Amazon is not rejecting agents as such. Amazon offers its own Buy for Me feature, which shops on external brands' sites, and emphasizes that it identifies itself and lets brands opt out. Walmart also blocks agents that violate its terms. The real difference between the two is whether the retailer provides an official entry point for agents that identify themselves and follow the rules.
| Item | Walmart (announced October 2026) | Amazon |
|---|---|---|
| Stance on outside agents | Accepts agents that follow its terms and provides a dedicated entry point | Blocks unauthorized agents as a violation of its Conditions of Use (blocked Muse in September 2026) |
| Connection path | MCP-based Agent Gateway (starting with a beta for companies and developers) | No public entry point for outside agents has been announced |
| Agent identification | Agent Profiles separate each agent's orders from the customer's own | Requires agents to identify themselves (says Muse does not) |
| Payment information | Agents do not receive card information and cannot change payment methods | Objects that Muse appears to store customer credentials |
| Mistaken purchases | Regular returns process applies, including when an agent misbehaves | No stated policy for mistaken purchases by outside agents |
| Own agent offerings | Sparky, plus integrations with ChatGPT, Gemini, and Muse | Buy for Me (identifies itself and, Amazon says, lets brands opt out) |
Consumers Are Not Yet Ready to Hand Over Control
Walmart's design is careful, but demand may not catch up right away. In a survey of 1,000 U.S. adults by payments provider NMI, only 11% had used AI to complete a transaction. Seventy percent said it is important to be able to review or override an AI decision before a purchase, and only 10% would give AI full control over a purchase.
Making approval the default fits these numbers. But if every purchase needs approval, the time saved by delegating to an agent shrinks. How widely customers use automatic purchasing within a spending limit will determine the real value of the system. And because what customers tell an agent's app is governed by the agent provider's policy, the boundary of responsibility remains hard for customers to see.
What This Means for Ecommerce Merchants
For Walmart Marketplace sellers, the shift has already begun. In March, Danker revealed that vitamin and protein supplements were the top two product types purchased from Walmart through ChatGPT.
Manish Joneja, senior vice president of Walmart Global Marketplace, told Digital Commerce 360 that agentic commerce could create opportunities even for brands without broad awareness. At the same time, he said accurate product information, competitive prices, availability, and reliable delivery promises will become increasingly important. The bar for being chosen by AI is stricter on information accuracy.
For merchants running their own ecommerce sites, Walmart's three-part framework doubles as a checklist.
- Entry point: Will you block agents or provide an official connection? Are your bot defenses turning away partner agents?
- Identification: Can you record and measure agent-placed orders separately from the customer's own orders?
- Responsibility: How will you handle an agent's mistaken purchase? Do your return terms explicitly cover agent-placed orders?
The first item is easy to overlook. TechCrunch cites the view that Cloudflare's change to its crawler defaults on September 15 contributed to blocking personal agents. Blocking agents without meaning to is exactly what happened at Walmart.
Conclusion
Walmart's announcement is a comprehensive response from the retailer's side, handling agent-led purchasing in three layers: entry point, permissions, and rules. While Amazon uses its terms to keep unauthorized agents out, Walmart opens a door to agents that identify themselves and follow the rules, and even absorbs the cost of mistaken purchases.
The next things to watch are which agents support Agent Profiles and which companies join the gateway beta. How many customers actually let agents buy within a spending limit will decide whether this design becomes the industry standard.
