By Puneet Gupta and Siddharth Shekhar Singh, Respectively professor of finance, IMT Ghaziabad, and associate professor of marketing, Indian School of Business, Hyderabad & Mohali
India’s digital public infrastructure (DPI)—Aadhaar for identity, Unified Payments Interface (UPI) for instant payments, account aggregator (AA) for consented data sharing, and open commerce networks such as ONDC (Open Network for Digital Commerce) built on Beckn—has already proven that open protocols, not closed platforms, can transform an economy. They worked because the rules were open, the interfaces were standardised, and participation was easy. The next wave of value will come from letting software act across these rails on our behalf. That requires a new kind of connective tissue for artificial intelligence (AI). It is emerging now in the form of the Model Context Protocol (MCP), and its importance to India’s DPI mirrors the role application programming interface (API) played for the early internet. APIs standardised request–response; MCP standardises discovery, permissions, and multi-step action—precisely what agentic AI needs to operate safely in the real world.
Leaders often ask why a new protocol is needed when APIs already exist for every service under the sun. The answer is that AI systems are not merely “calling” a service; they are collaborating with many services, in sequence, with context. A customer-support agent that must look up an ONDC order, verify a refund policy, initiate a UPI payment, and notify a logistics partner is executing a chain of moves, each dependent on identity, consent, and auditability. In the API world, enterprises build bespoke connectors and glue code for each step, replicating governance logic dozens of times. In an MCP world, the agent can discover the tools it is allowed to use, negotiate the right scopes, carry forward state, and emit structured telemetry as it works—one portable pattern rather than a spaghetti bowl of integrations.
This has resonance in India because DPI already encodes the hard parts of trust. Aadhaar and e-KYC make identity programmable. The AA ecosystem gives firms a rulebook and artefacts for consent, purpose limitation, and revocation. ONDC/Beckn formalises how buyers and sellers find each other and transact without being trapped in a single marketplace. Open Credit Enablement Network (OCEN) does the same for credit flows to small businesses. MCP takes these building blocks and makes them actionable for AI, turning protocols that move data and value into workflows that move outcomes. An underwriting copilot can request consented cash-flow data through AA, compute risk, propose terms over OCEN, and disburse through UPI—with every step logged, reversible where needed, and visible to risk functions. A retail service agent can trace an ONDC order, check inventory, schedule a replacement, and initiate a refund—all within the boundaries set by policy rather than the creativity of a developer’s one-off script.
For CEOs and boards, the shift to MCP is less about another piece of technical plumbing and more about cycle-time compression and governance by design. When DPI capabilities are exposed once behind MCP, they can be reused across dozens of AI use cases without re-verifying the same security and compliance assumptions. Time-to-value shortens, not because the model is better, but because the surrounding affordances—discovery, permissions, logging, redaction, human-in-the-loop checkpoints—are standardised. At the same time, risk teams gain a coherent control plane: who or what called which capability, with what inputs and outputs, and to what business effect. In a regulatory environment that rightly emphasises consent, lineage, and auditability, MCP allows firms to move fast and stay exam-ready.
The managerial implications are straightforward and urgent. Treat MCP as a first-class standard alongside your India Stack integrations, not as an experimental add-on. Name an accountable product owner for your “AI railhead”—the team charged with exposing UPI, AA, ONDC/Beckn, and OCEN capabilities behind MCP with least-privilege scopes and rich telemetry. Fund flagship workflows that traverse multiple rails so you can measure end-to-end economics rather than isolated model metrics: onboarding an micro, small, and medium enterprise with consented data, embedded underwriting, and instant disbursal; or automating returns and replacements with transparent refunds and predictable logistics. Build an internal catalogue that makes these capabilities discoverable to every sanctioned copilot and agent, and insist that new use cases consume the standard rather than reinvent it.
Sceptics will argue that open protocols are slower or less secure than closed, vertically integrated solutions. India’s experience with DPI offers a practical rejoinder. Open protocols, paired with strong rulebooks, unlock innovation while raising the floor on trust. MCP carries that ethos into AI. It does not eliminate the need for human oversight; it makes oversight feasible at scale. It does not pick a winning model; it future-proofs your integration layer against inevitable model churn. Most importantly, it reframes AI from a feature bolted onto apps into a fabric that weaves through identity, payments, data, and commerce—exactly where India has already built durable advantages.
The strategic prize is twofold. First, firms that adopt MCP atop DPI will convert AI from a cottage industry of pilots into an operating system for growth: faster service, lower unit cost, and new routes to market. Second, India can export not just digital rails, but also a governance pattern for AI-mediated action—one that balances openness with enforceable consent and verifiable outcomes. APIs made the web programmable. India’s DPI made a nation interoperable. MCP can make both assets actionable together, so that AI in India is not merely a way to answer questions, but a dependable way to get things done.
