Social Rocket
DoorDash is opening its ordering processes more widely to AI agents and chat interfaces. For businesses, the delivery service is testing its own Model Context Protocol (MCP) server, through which authorized agents can search restaurants and merchants, assemble shopping carts and prepare orders. In parallel, a pilot is running in the US in which consumers can message DoorDash via Apple Messages and, for example, reorder previous orders.
Adweek reports on 30 September 2026 that the messaging pilot has already been enabled for around 20.000 US users. The business features, by contrast, are still in a limited beta.
DoorDash MCP brings orders into work tools
The more significant infrastructure development is DoorDash's official MCP server for businesses.
DoorDash describes typical use cases such as recurring office orders, reorders of consumables, or group orders that can be coordinated via a Slack bot, for example. An authorized agent can search restaurants and retail offers, create a shopping cart, manage delivery details and query the status of an order.
The order remains tied to a specific user. DoorDash uses OAuth for authorization; according to the documentation, the agent should not see raw credentials. Before an order is finally submitted, DoorDash also requires explicit confirmation from the end user.
MCP access is currently in private beta and, according to DoorDash, is not intended as a freely available interface for arbitrary consumer-facing third-party products.
From text message to shopping cart
DoorDash is also testing a much simpler consumer interface.
According to Adweek, selected iPhone users can send the service requests or a message such as "my usual order" via Apple Messages. The assistant searches for suitable options and guides the user through the ordering process.
This moves commerce from a classic app interface into a conversation. We have seen this pattern across several areas by now: Meta, OpenAI, merchants and delivery services are all trying to shift product search, recommendation and checkout into agentic or conversational interfaces.
We have already shown how Kroger and DoorDash are integrating AI more deeply into the shopping funnel. Meta Muse is also becoming a test case for agentic commerce, while Google is bringing merchants closer to such systems with new AI and commerce features in Merchant Center.
Why Slack and MCP matter for marketers
The decisive point is less whether office lunches will actually be ordered in a Slack channel in future. More interesting is the new interface between intent and commerce.
If a user no longer opens a merchant app first, but instead gives an agent a task, the classic customer journey changes. Product data, availability, price, recommendations and brand preference then have to work in a machine-readable environment.
For retail media, this also raises a measurement question. DoorDash is already building its advertising business on purchase and order data and is investing in more independent impact measurement. Our articles on DoorDash's certified incrementality measurement and on shelf and inventory data at DoorDash and Instacart show how closely commerce infrastructure and media are now converging.
Agentic commerce gets real transactions
Many AI commerce demos still end at a product recommendation or a link to the merchant. With MCP, DoorDash goes a step further: within an authorized workflow, the agent may build a shopping cart and, after user confirmation, submit an order.
That is exactly why the confirmation logic matters. As soon as an agent can trigger a transaction, permissions, identity and traceable consent become core components of the product.
For marketers, this means agentic commerce is no longer just a new search or recommendation format. The next phase concerns transaction, attribution and access to the customer.