Swiggy MCP: 3 Official Servers, 35 Tools, Food to Instamart to Dineout

Swiggy runs its official MCP integration through what it calls the Swiggy Builders Club — and unlike most single-server brand integrations, Swiggy actually ships three independent MCP servers, one per business line, rather than one server trying to cover everything.

Three Servers, Not One

• Food (mcp.swiggy.com/food) — 14 tools covering restaurant discovery, menus, ordering, and order tracking.
• Instamart (mcp.swiggy.com/im) — 13 tools for quick-commerce grocery shopping.
• Dineout (mcp.swiggy.com/dineout) — 8 tools for restaurant table reservations.

That's 35 tools total across the three servers — but they are deliberately siloed: carts, orders, and sessions are not shared between Food, Instamart, and Dineout. An AI agent talking to the Food server has no visibility into an Instamart cart, and vice versa.

Who It's Built For

Swiggy's documentation names three real audiences:

• Agent developers building on OpenAI, Anthropic, LangGraph, or Vercel AI SDKs
• Agent platform operators running things like voice assistants or in-app agents at scale
• Individual users connecting Swiggy directly to Claude Desktop, ChatGPT, Cursor, or VS Code

Authentication is OAuth 2.1 with PKCE over standard JSON-RPC — a genuinely modern, spec-compliant setup rather than a bolted-on API key.

The Honest Limitations

Two real constraints are worth knowing before you build against this:

• India-only, for now. Access is currently restricted to India-based users.
• Production access is gated. Moving beyond a personal/testing setup to production scale requires whitelist onboarding and a partner contract with Swiggy — this isn't an open, self-serve API key signup.

Separately, reporting from MediaNama has noted that ordering via ChatGPT on Swiggy hasn't always worked smoothly end-to-end in practice, particularly on the Instamart side, where grocery checkout has at times fallen back to the regular app rather than completing fully inside the AI chat. That's a useful reminder that "official MCP server exists" and "every flow works flawlessly through every AI client" are two different claims — worth testing directly before relying on it for anything production-critical.

Why Three Servers Instead of One

Splitting Food, Instamart, and Dineout into separate MCP servers with separate tool sets and separate sessions is a deliberate architectural choice, and a reasonable one: it keeps each server's tool list small and specific rather than one bloated 35-tool server trying to do everything, and it means a client that only cares about restaurant reservations (Dineout) never has to reason about grocery-cart tools it'll never use.