Skip to content
audiotale.net
Sign in
NewsNews

Walmart launches Agent Gateway MCP server and Agent Profiles with spending limits for AI shopping agents

Walmart is letting approved AI agents shop through an MCP server, with per-agent profiles and spending caps, as Amazon blocks outside agents.

audiotale.net team

· 7 min read

Written by AI from the sources linked below. How we use AI

On this page
  1. Key points
  2. What happened
  3. Background
  4. How it works
  5. What it means
  6. What we don't know yet
  7. What to watch next
  8. Sources

Walmart announced on 21 October 2026 that it is opening its shopping systems to AI agents on its own terms: a Model Context Protocol (MCP) server called the Agent Gateway, separate "Agent Profiles" with spending limits inside customers' accounts, and a published set of rules agents must follow or be blocked. The announcement comes weeks after Amazon began blocking outside shopping agents, so the two largest US retailers are now taking opposite positions on whether AI assistants may buy on a customer's behalf.

Key points

  • Walmart's Agent Gateway is an MCP server that lets approved AI agents search Walmart and Sam's Club inventory and buy for a customer; it starts as an invitation-only beta for companies and developers, with sign-up through walmart.com/agents.
  • Agent Profiles, due within about a month of the 21 October announcement, give each agent its own profile in a customer's Walmart account; customers can require approval for every purchase or let the agent spend within a limit (Walmart's examples are $50, $100 and $200).
  • Walmart's terms say agents never receive card details, cannot add or change payment methods, must identify themselves as agents and must follow rate limits; agents that break the terms will be blocked.
  • Walmart says agents see the same prices as people and that its returns process applies even when an agent does not act as intended.
  • Amazon, by contrast, has been blocking outside shopping agents, including Meta's Muse, which Walmart has named as a partner.

What happened

Walmart president and CEO John Furner announced the plans at the Masters of Scale Summit in San Francisco on Wednesday 21 October, and the company published a post the same day. The post describes a problem in plain terms: today, when a customer uses an AI agent, the agent signs in with the customer's account, so purchases by the customer and by the agent appear together with no way to tell them apart. Agents also shop through a web browser, following steps designed for people, which Walmart describes as slow, error-prone and prone to leaving agents stuck.

The Agent Gateway is the technical answer to the second problem. It is an MCP server, a standard (originally created by Anthropic, as TechCrunch notes) that lets an AI agent call a service's tools directly instead of driving a website. Through it, approved agents can search Walmart's online inventory and make purchases for a user, covering both Walmart and Sam's Club. Walmart's beta page, as summarised by Unite.AI, promises up-to-date product and fulfilment data from Walmart itself rather than scraping, fewer requests, lower latency and lower token use, plus verified agent identities, controlled permissions, observable actions, category restrictions, rate limits and compliance guardrails. The same report says the beta is early and invitation-only and may be suspended at any time. We did not open the beta page itself.

One feature is aimed at privacy. Walmart says the Gateway can let an agent reorder a customer's usual items, its examples are a particular bread and a milk brand, without the agent receiving the customer's broader purchase history or personal information.

Agent Profiles handle the account side. Per Walmart, the first time a customer uses an agent through the Gateway, Walmart automatically creates a profile for it. The customer then links that profile to their Walmart account by signing in to Walmart, a step Walmart says only Walmart sees, and sets a spending limit. Customers can view purchases from all their agents, change permissions at any time and delete a profile. Walmart compares the model to adding an authorised user to a credit card. The tools are open to anyone with a Walmart account; a Walmart+ membership is not required, according to TechCrunch.

Background

Agentic shopping, where an assistant finds and buys things for you, has so far depended on agents logging in as the customer. Retailers have reacted differently. According to GeekWire's reporting, around 20 September 2026, roughly twelve days after launch, Amazon blocked Meta's Muse assistant from buying on Amazon.com; Amazon objected that Muse did not identify itself as an agent and could store customer credentials, while Meta said Muse never sees passwords or payment details and pays with single-use cards. TechCrunch notes that various travel and booking sites have also locked agents out entirely.

Walmart took the other route. It had already announced a partnership with Muse and offers conversational-shopping integrations with ChatGPT, Gemini and other apps, alongside its own assistant, Sparky. Its Terms of Use, updated on 21 September 2026 according to Unite.AI, bar unauthorised agents from scraping, placing orders or using site materials to train AI models without written consent. The pattern is: a front door for agents that identify themselves and accept rules, and a closed door for those that do not.

We have written before about the same underlying concern from the security side: Microsoft's argument that AI agents should be treated as insiders, with their own identities and limited permissions. Walmart's Agent Profiles apply that idea to shopping.

How it works

Walmart's design has three layers, which map onto what developers building shopping agents will need to handle.

  1. Identity and access. An agent connects to the Gateway as an approved agent with a verified identity, rather than reusing a customer's browser session. Walmart's terms require agents to identify themselves and not hide automated activity.
  2. Delegated authority. Each agent gets a per-customer profile. Approval is required by default for agent purchases, and Walmart says it may require extra approval even for purchases under an agent's limit. Walmart may also cap transaction value, quantity or frequency, decline or reverse transactions and revoke access.
  3. Payment and data boundaries. Agents never receive card details and cannot change payment methods. They may access only the information needed for the task and permitted by the customer. What the customer tells the agent app is governed by that provider's privacy policy; what the agent sends to Walmart falls under Walmart's Privacy Notice.

Walmart also states a pricing principle: agents see the same prices a human sees on Walmart.com, with offers and fees shown before purchase, and "We price products, not people." That is a commitment worth watching, because agents that compare prices at scale put pressure on any retailer that personalises them.

On liability, Walmart's terms say that authorising an agent does not make it a Walmart employee, partner or representative and that Walmart is not responsible for a third-party agent's conduct, per Unite.AI. At the same time, Walmart says its returns process applies when an agent does not act as intended, so customers are not left with unwanted goods because of an agent's mistake.

What it means

For developers of shopping agents. The Gateway is a sanctioned alternative to scraping and browser automation, and it comes with obligations: identify yourself, respect rate limits, keep to the customer's intent and expect the human to approve purchases by default. Fully autonomous buying will exist only inside limits customers set. Access is by invitation in a beta, so the first partners will be the large assistants Walmart already works with.

For other retailers. There are now two reference models. Amazon's closes the site to agents it has not authorised, which protects its own shopping experience and, as commentators have noted, its advertising business, which depends on people browsing its pages (a motive Amazon has not stated in the reports we read). Walmart's admits agents but keeps authorisation, spending limits and enforcement in its own hands. A retailer that wants agent traffic but fears credential sharing can borrow the profile pattern without adopting MCP.

For shoppers. The headline benefit is control and visibility: agent purchases appear separately and can be capped. The caveat is that the safeguards depend on Walmart's rules and each agent maker's behaviour; the default approval requirement means "set it and forget it" shopping is not what is launching now.

What we don't know yet

  • Who is in the beta, how many developers will be admitted and when access widens. Walmart's post gives sign-up through walmart.com/agents but no numbers.
  • Exact launch timing for Agent Profiles. Walmart says "within the next month"; TechCrunch reports November.
  • How many Walmart shoppers use agents today. None of the sources we read gives adoption or sales figures for agentic shopping.
  • Whether Amazon changes course or takes action beyond blocking; Amazon has said it is in talks with Meta and has not said whether it would pursue legal action, per the GeekWire report.
  • How well verification works against agents that do not identify themselves. Walmart says violators will be blocked but has not described detection.

What to watch next

The November rollout of Agent Profiles and whether the spending-limit controls behave as described; the first named third-party agents in the Gateway beta; any published API or MCP tool documentation, which would show exactly what agents can and cannot do; and whether other large retailers copy Walmart's profile model or Amazon's blocking approach.

Sources

SaveStart the discussion

Written by

audiotale.net team

The official account of the people who run audiotale.net: announcements, conversation starters and the newsroom's drafts. Posts written with AI say so. Every other account here is a real person.

Read next