Zepto, Swiggy Instamart and Blinkit MCP: Official Status

Zepto MCP and Swiggy Instamart MCP have published official connection instructions. Blinkit is different: we did not verify a company-published MCP endpoint in this review. For Indian quick commerce, distinguish documented access from community automation before connecting an account or designing a purchasing agent.

Reviewed 23 September 2026. This is a documentation review, not an authenticated checkout test. Availability can depend on the client, account, delivery location and production-access approval.

Correction: The previous version treated these integrations as hypothetical and forecast future launches. That was incorrect for Swiggy and Zepto. This revision replaces those forecasts with primary-source documentation, removes unsupported order-volume and performance claims, and does not infer Blinkit capabilities from another company's integration.

Which quick-commerce MCP integrations are documented?

Sources: Zepto's MCP repository, Swiggy's server reference, and the Swiggy developer quickstart, reviewed September 2026. The Blinkit status describes the limits of our verification, not proof that no private or future integration exists.

Zepto MCP: where to start

Zepto's repository documents product search, cart changes, ordering and order history through its MCP service. It describes an India customer integration authenticated through OAuth and a phone verification flow. Its payment section lists COD, UPI, cards, Zepto Cash and UPI Reserve Pay. Treat these as documented options, not a guarantee that every order offers all of them.

Use the endpoint in the table and follow the client-specific instructions in the official Zepto README. Do not substitute a similarly named package or repository without checking its maintainer and authentication destination. The README distinguishes MCP connectivity from broad app availability; support in one client does not establish availability in every assistant.

Zepto explicitly warns that orders placed through this integration are real. Start with a read-only product lookup, then inspect the cart. Do not use an order-placement request as a connectivity test. We did not authenticate a customer account, place an order or verify delivery during this review.

Swiggy Instamart MCP: use the current endpoint

Swiggy announced MCP integrations for Food, Instamart and Dineout on 27 January 2026. Its launch announcement links an Instamart URL ending in /instamart. The current server reference and official repository instead show /im. Use the current reference when configuring a client; this review did not test whether the older path redirects.

Swiggy's developer quickstart separates local development from production approval and describes OAuth authentication. Its platform overview treats the service endpoints as independent: a food cart is not an Instamart cart. Do not design a shared basket on the assumption that a shared brand means a shared transaction.

Keep grocery and restaurant workflows separate in your agent, including the confirmation shown to the shopper. Before launch, check the current service-specific payment restrictions and access conditions in Swiggy's documentation. A successful product search does not establish permission or readiness to place an order. For the wider food-delivery context, see our Zomato and Swiggy MCP guide.

Blinkit MCP: official support is not verified

Searches for Blinkit MCP return community implementations. For example, hereisSwapnil/blinkit-mcp explicitly describes itself as unofficial and disclaims authorization or endorsement by Blinkit. That is primary evidence about the project's affiliation, not evidence of a Blinkit-operated developer service.

We did not find a company-published Blinkit MCP endpoint in the Blinkit and Eternal sources checked on 23 September 2026. Do not relabel a community wrapper as an official connector, and do not assume a Zomato integration extends to Blinkit. For a production dependency, ask the platform for written access conditions and support ownership before committing to an implementation.

An experimental wrapper can demonstrate a possible workflow without establishing a stable business integration. Review how it handles customer sessions, permissions and checkout. The existence of code on GitHub alone does not answer those questions. This article does not recommend installing an unofficial checkout automation tool or providing one with account credentials.

A practical evaluation sequence

The following is our implementation checklist, not a claim about a platform's internal ranking system or a report of tests we ran. Use it to separate discovery, authorization and ordering instead of treating a fluent conversation as proof that the complete workflow works.

1. Verify the publisher. Start with the platform's own documentation and follow its repository or endpoint links. Record the URL and review date. Check both the MCP host and the authentication host before signing in. A familiar logo in a client is not enough.
2. Establish the delivery location. Ask for location before promising availability. Product names alone do not establish local stock, price or serviceability. Use a deliberately unavailable item in your approved test environment to check whether the agent admits the limitation instead of inventing a substitute.
3. Keep the first call read-only. Search for a product and examine its identifiers, size, availability and price. Confirm which fields actually come back. Do not let a free-text product description override the shopper's constraints or the agent's tool permissions.
4. Review the basket before purchase. Show quantities, variants, substitutions, fees, address and total. Require a separate confirmation before the irreversible action. A request to find dinner ingredients is not automatically permission to buy every ingredient the agent chooses.
5. Handle uncertain outcomes. If a checkout call times out, establish whether an order exists before retrying. Preserve the platform's order identifier and show the customer the actual outcome. A network error after submission is not proof that nothing was purchased.
6. Separate access from reliability. Log unsuccessful searches, stock changes, authentication failures and checkout errors separately. A working demo tells you little about repeated use. Agree who handles incidents and how the shopper returns to the platform's normal support flow.

Our inventory-accuracy guide explains location and stock checks in more detail. The assistant deployment checklist covers ownership, catalog readiness and operational responsibilities.

What brands should prepare

An MCP connection does not itself guarantee that an agent recommends your products. It exposes a set of capabilities; the shopper's constraints, available catalog data and the agent's decisions still matter. We have not established a preferred-brand ranking formula for any of these platforms, and we do not claim that early integration buys preferential traffic.

Prepare product attributes that make comparisons unambiguous: pack size, ingredients, dietary constraints, variant identifiers and permitted substitutions. Keep availability and price tied to the actual delivery context. Track product discovery, cart additions and completed orders separately so that an improvement in one is not presented as an improvement in all of them.

For the catalog work, start with product data for AI agents. For an implementation on a commerce system you control, review the MCP app builder. Our Amazon case study is a separate account of measured visibility work, not evidence of results from Zepto, Swiggy or Blinkit MCP.

Use this guide as a current access map, then validate the exact workflow you intend to support. Keep a dated record of the source documentation and revisit it before a production rollout. Return to the implementation guides for the broader commerce workflow, or discuss an MCP implementation with Tru Commerce.

Related: Meta Muse is not in India yet

Meta's Muse agent launched in the US on 8 September 2026 and reached Canada on 18 September; Meta has named no India date and runs no waitlist. What an India launch would meet, from CERT-In's human-in-the-loop line to NPCI's Unified Agent Protocol being pulled pending RBI clearance, is in Meta Muse in India: when it lands, what it will meet.