Key Takeaways
- Swiggy unveiled Builders Club, an invite-only programme that opens three MCP servers, 18+ API tools and reusable "Skills" to external developers, startups and enterprises.
- The move makes Swiggy Food, Instamart and Dineout addressable by AI agents, repositioning the company from a consumer app into what its CTO calls an "ecosystem orchestrator."
- For e-commerce operators, it concretises a question that Western platforms are still debating: which slice of your commerce domain do you expose to agents, and through which interface?
Swiggy opens its AI commerce stack with Builders Club
Invite-only initiative offers APIs, MCP servers and 'Skills' so startups and enterprises can create real-world AI commerce integrations on Swiggy
www.fortuneindia.comOn April 23, 2026, India's on-demand commerce major Swiggy announced Builders Club, an invite-only programme for developers, startups and enterprises. The initiative opens the transactional surfaces of Swiggy Food, Instamart (quick commerce) and Dineout (restaurant reservations) so external teams can embed them into AI agents and copilots.
At launch, the programme exposes three MCP servers, over 18 API tools and reusable agent capabilities called Skills. Approved builders can compose these into agents that order meals, buy groceries or book dining experiences, not merely recommend them. The distinction matters: Builders Club targets the transaction itself, not discovery.
CTO Madhusudhan Rao framed the shift in explicit language: "Builders Club is the next bold step, extending that access to developers and enterprises so they can build AI commerce applications at scale on top of Swiggy. We are moving from platform to ecosystem orchestrator."
What is actually inside Builders Club
The architecture splits into three layers. The infrastructure layer sits on AWS, with Amazon Bedrock offering unified API access to foundation models from Anthropic, Meta, Mistral AI and others. AWS Trainium underpins training and inference, delivering what Swiggy describes as up to 50% training-cost savings and 30 to 40% inference-performance gains. Amazon Bedrock AgentCore handles the agent-execution framework.
The middle layer is the MCP server set. Model Context Protocol, the Anthropic-originated spec now moving toward industry standardisation, defines how LLMs access external tools and data safely. Swiggy had already rolled out a more limited MCP surface; Builders Club sits on top as the ecosystem layer that determines who builds on the infrastructure and how.
Skills crown the stack. Swiggy defines them as reusable capabilities that let AI agents complete real-world tasks. The concept aligns with Anthropic's Agent Skills framing: bundle several API calls into a use-case-level module. A single Skill might handle the full "suggest dinner, confirm order, arrange delivery" loop, shielding agent developers from the underlying plumbing.
Who can join and what they get
Participation is invite-led. Developers, startups and enterprises apply, pass review, and then get live API access. Support goes beyond API keys: selected participants receive engineering assistance from Swiggy's team, co-branding opportunities, and what the company calls growth partnerships that elevate promising use cases into formal commercial arrangements. The path from sandbox demo to production transaction is explicitly engineered into the programme.
Sandeep Dutta, President of AWS India and South Asia, positioned the initiative as AWS's showcase for AgentCore at commercial scale, saying the combination of Bedrock, Trainium and AgentCore gives builders "infrastructure to innovate without limits." Swiggy gets an enterprise-grade backbone; AWS gets a reference customer for agentic workloads.
How Swiggy fits against the global field
How commerce companies open themselves to agents has become the defining 2026 question. Setting Builders Club alongside other platforms clarifies Swiggy's stance.
| Platform | Commerce domain opened | Agent-facing interface | Monetization direction |
|---|---|---|---|
| Swiggy Builders Club | Food, Instamart (quick commerce), Dineout | 3 MCP servers + 18+ API tools + Skills | Commerce take-rate through invite-led ecosystem |
| Shopify (Catalog / Commerce API) | Merchant-owned storefronts | REST/GraphQL API, Agentic Commerce partner integrations | Merchant subscription + transaction fees |
| Stripe (Agent Toolkit) | Payments, Issuing, Connect | Agent Toolkit SDK + MCP | Payment fees (same whether agent-driven or not) |
| Zomato (domestic competitor) | Food primarily | No public MCP / Builders Club equivalent yet | Traditional in-app commerce |
Shopify is opening merchant storefronts to agents through its Agentic Commerce partner programme. Stripe is pitching Payments, Issuing and Connect as the neutral payment rails agents plug into. Swiggy is different: it owns the supply side (restaurants, groceries, dining venues) and is handing that supply to agents wholesale through APIs. It is a vertically-integrated agentic commerce play, not a horizontal toolkit.
Domestic rival Zomato has not announced a comparable developer-ecosystem programme. By moving first, Swiggy is trying to shift the competition from app-versus-app into infrastructure-versus-infrastructure, a frame where its scale and supply network count more than its consumer brand.
Implications for India's e-commerce ecosystem
India's mobile economy is expanding fast, but device and network diversity cap how many native apps any user can sustain. That makes agent-mediated commerce a more plausible growth path than in Japan or the West. If WhatsApp or third-party AI assistants become the de facto shopping entry point, Swiggy can reach users who never installed its app.
The move also aligns with the revenue-mix pressure Swiggy faces post-IPO. Relying on delivery margins is fragile; platform fees, API-based transaction fees and expanded GMV routed through external agents diversify the base. The phrase "foundational layer for AI-native commerce innovation" reads as a deliberate repositioning from demand-capture app to distribution-aggregation infrastructure.
The pull on the developer side is real as well. Nowhere else in Indian food, quick commerce or dining can a builder obtain sanctioned APIs plus transaction-capable MCP servers in one package. Search-style assistants, budget and meal-planning apps, corporate benefits platforms and travel copilots all have room to layer value on top.
What global e-commerce operators should take away
This is geographically an India story but strategically a universal one. The first honest question is whether your own products and services can be invoked by an AI agent at all, and at what abstraction level.
Interface design is a product responsibility. Swiggy did not merely publish APIs — it published three MCP servers, 18+ tools and higher-order Skills. That reflects a product organisation that chose the granularity at which to hand its domain to agents. Shipping "stock and order APIs" and stopping there cedes use-case ground to competitors who ship Skills.
Invite-led governance is becoming the template. Fully open APIs invite abuse and malicious agents; fully closed ones starve the ecosystem. Application, review, tiered access and performance-based partner elevation look increasingly like the standard operating model for agentic-commerce platforms.
Identity, payments and trust warrant explicit choices. Swiggy picked AWS for infrastructure and Bedrock for multi-model flexibility, betting on lock-in at transaction completion rather than at the model. Every platform planning an agent surface must decide which layers to differentiate in and which to absorb from the standard stack.
One risk deserves explicit mention. As agent-routed transactions grow, the revenue tied to in-app UI, recommendations and ad surfaces contracts. Any Japanese or Western operator that leans heavily on app-based advertising and affiliate revenue should already be modelling that shift.
Conclusion
Swiggy's Builders Club moves the "AI commerce" conversation from consumer chat experiences to infrastructure design on the platform side. Three MCP servers, 18+ API tools and Skills are a concrete instantiation of abstract concepts that Anthropic and Google have been promoting — this time bolted onto the messy realities of food, groceries and dining.
What to watch next is how Zomato, Amazon and Flipkart respond, and whether similar openness emerges in adjacent verticals. The agentic-commerce contest is shifting away from model quality and chat UI toward who assembles executable APIs, MCP servers and Skills first. Operators globally, including in Japan, should use 2026 to audit what it would take to hand their catalogue, inventory and checkout to agents at first-party grade.
