AI NewsDev toolsReported

Walmart opened an MCP server so AI shopping agents can buy from it

Walmart announced an Agent Gateway, an MCP server that lets approved AI agents search its catalog and place orders, and a separate Agent Profiles feature coming next month that gives each agent its own account and spending limits.

AI News

Editorial3 min read

LinkedInX
TechCrunch header photo of a Walmart parking lot used on the Agent Gateway story

Image: TechCrunch

Why it mattersWalmart adding a sanctioned path for AI agents to shop while Amazon blocks them puts the Model Context Protocol in a position to decide which retailers end up on an agent's shortlist for a long time.

Walmart has built a front door that AI shopping agents can walk through. TechCrunch reports that the retailer announced an Agent Gateway, an MCP server, on 21 October at the Masters of Scale Summit in San Francisco. The gateway lets approved AI agents search Walmart's inventory and place orders on behalf of a customer who signed into the agent. Walmart Chief Executive John Furner is cited as the speaker.

TechCrunch names three agents that work with it today: Meta's Muse, ChatGPT and Google's Gemini. The gateway covers Walmart and Sam's Club. The timing is pointed. Amazon moved in the other direction this week and started blocking AI agents from making purchases on its own site, which TechCrunch frames as a contrast with Walmart's open posture.

What the Agent Gateway actually does

The piece describes the gateway as an MCP server. MCP, the Model Context Protocol, is the way agents built on top of ChatGPT, Claude and Gemini discover and call external tools. By running one, Walmart gives any agent that speaks MCP a sanctioned method to look up what Walmart sells, check prices and place an order, without pretending to be a browser or scraping pages. The agent can see what the catalog shows a human and act on it inside the limits Walmart sets.

Walmart's terms, as TechCrunch summarises them, say agents cannot access a customer's payment information, cannot take unintended actions and cannot ask for more information than the task needs. Standard returns policies apply even when an agent misbehaves. Those three lines are a reasonable baseline for consumer safety and leave a lot of practical questions open.

Agent Profiles, coming next month

The separate piece in the announcement is Agent Profiles, which TechCrunch dates to a rollout next month. Each customer can authorize an individual agent with its own account profile, see its purchases separately from their own, set a spending limit and require approval on each transaction. Compared with an open-ended token that an agent uses to shop as the human, this is closer to a corporate credit card issued to a vendor: separate line items, a cap, and a review on the ones above the cap.

For any team building an agent that pays for things, that pattern is worth looking at. The person who signs in to the agent keeps the human account, and the agent gets a separate shadow account with its own history and its own limit. If something goes wrong, the audit trail is already split.

What is not clear

The piece does not state how an agent is approved to use the gateway, which the three named ones have already cleared. It does not say whether a smaller agent runtime, say a custom agent built by a team on top of an open model, can register for the gateway or must wait. It also does not include a rate limit, a request format or a schema reference from Walmart's engineering side; those will show up when the retailer publishes developer documentation, if it does.

The commerce surface each agent can reach is becoming a competitive layer. One retailer welcomes agents, another blocks them, and the agents will develop a preference. Walmart is making the first move.

Source

TechCrunch: While Amazon blocks AI agents, Walmart rolls out the welcome mat.

Reported byTechCrunch

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project