Zomato MCP and Swiggy MCP: India's Apps Inside AI Assistants

Zomato and Swiggy both run MCP servers, so an AI assistant can search restaurants, build a cart and place an order without the app ever opening. Zomato shipped an official open-source server in September 2025; Swiggy followed in January 2026 across food delivery, Instamart and Dineout. Between them they are the clearest working example of agentic commerce in India — and the template every other Indian brand is about to be measured against.

Most explanations of the Model Context Protocol are written for people building developer tools. That is not where MCP got interesting in India. It got interesting when you could type "order me a masala dosa" into Claude and a real order appeared at a real restaurant.

Two companies made that true. Their approaches differ enough to be worth reading side by side, especially if you are a brand wondering what your own version looks like.

Zomato: an official, open-source server since September 2025

Zomato publishes an official MCP server. The manifest lives at github.com/zomato/mcp-server-manifest, it registers under the server name com.zomato/mcp, and it was released on 24 September 2025.

The details that matter if you are evaluating it:

• Remote, not local. It is a hosted endpoint using streamable HTTP transport, current with the MCP specification. You are not running a process on your laptop.
• OAuth, not API keys. Connecting triggers an OAuth flow. There is no key to paste, and no key to leak — which is the correct answer for a server that can spend a user's money.
• End to end, not just search. It discovers restaurants, browses menus, builds a cart, applies offers, and completes the order. That last step is what separates a real agentic integration from a lookup tool.
• Free, with the manifest carrying its own installation and configuration instructions.

The scope is deliberately narrow: search restaurants, order food. It does one job and finishes it.

Swiggy: three verticals at once, January 2026

Swiggy went wider. As Inc42 reported on 27 January 2026, Swiggy brought MCP to three businesses simultaneously:

It works across ChatGPT, Claude, Gemini and Cursor. Swiggy's CTO Madhusudhan Rao framed the intent as removing friction from daily decisions — "enabling a level of ease, personalisation, and joy that makes on-demand convenience feel effortless."

Swiggy claims Instamart as the first quick-commerce platform anywhere running on MCP.

One honest caveat. Shortly after launch, MediaNama reported that ordering through ChatGPT on Swiggy's services was not working reliably. Early agentic integrations are genuinely hard, and the gap between "the endpoint exists" and "a stranger can complete a purchase through it" is where most of the engineering actually sits. Worth knowing before you promise your own launch date.

If you searched "Instamart MCP" or "BigBasket MCP"

A quick clarification, because these are common searches and the answer is not obvious.

Instamart does not have a separate MCP server. It is one of the three verticals inside Swiggy's MCP integration. If you are looking for Instamart, you are looking for Swiggy's.

BigBasket took a different route. As Mint reported, the Tata-owned grocer enabled users to shop and pay inside ChatGPT — an app-surface integration rather than a published MCP server. Same destination, different road.

Zepto's started as an internal experiment by one of its own engineers, was published to GitHub, and was then promoted by co-founder and CEO Aadit Palicha on LinkedIn. That origin story is worth noting: the first version did not come from a roadmap.

What this pattern actually tells brands

Four things are now clear from watching India's largest consumer platforms do this in public.

1. Transaction beats lookup. Every one of these integrations goes all the way to a completed order. A server that only answers questions gets used once. A server that completes a purchase becomes a channel. This is the whole difference between being discoverable and being transactable.

2. OAuth is the pattern, not API keys. Once an agent can spend money, credential handling stops being a convenience question. Zomato's choice here is the one to copy.

3. Breadth is a product decision, not a technical one. Zomato shipped one vertical properly. Swiggy shipped three at once and took reliability criticism for it. Neither is wrong — but pick deliberately rather than by accident.

4. Nobody is waiting for a standard to settle. ACP, UCP, AP2 and MCP are all still moving. These companies shipped anyway, and are learning things their competitors will read about instead of discovering. If you are waiting for the protocol war to resolve before you start, you are choosing to learn last. We wrote up how the protocols actually differ if you need to pick one.

Where this leaves everyone else

Zomato and Swiggy are food and grocery. The same architecture applies to any catalog with a checkout — beauty, apparel, electronics, pharmacy. The reason it showed up in food first is that food has the highest order frequency and the lowest deliberation, so the payoff from removing friction is largest and arrives fastest.

That is a sequencing observation, not a moat. The infrastructure is identical.

For most brands the honest constraint is not the protocol. It is that the catalog is not machine-readable, the checkout cannot be driven by anything that is not a browser, and nobody owns the problem internally. That is the work — and it is the work we did with House of Zelena, where the outcome was not a dashboard number but which conversations the brand could actually complete a sale inside.

If you want the same thing pointed at your own catalog, that is what our MCP App Builder exists to do.

The window here is narrower than it looks. Right now, being the only brand in your category that an assistant can actually buy from is a real advantage. It stops being one the moment your competitor ships their server — and in India, two of the largest consumer platforms have already shown everyone exactly how.