Why the Model Context Protocol (MCP) is a pragmatic bridge, not a moonshot rewrite
We’re at that point in the AI hype cycle where two very different product strategies are getting lumped together: one aims to empower users to do more with AI, the other is about making the product smarter. Both are valid, but they’re not the same - and they lead to very different decisions about infrastructure.
For most enterprises, the smartest short-term strategy isn’t to rebuild. It’s to extend. That means fronting existing API investments - not rewriting them. It means deploying MCP servers that call existing APIs as-is, adding just enough context and tooling to make them usable by LLMs, agentic runtimes, and next-gen orchestration frameworks.
Don’t start with a new backend. Start with a proxy. on existing services
That’s what MCP unlocks. Not a new protocol to replace your APIs - but a contextual layer that helps external agents understand what your APIs do, how they should be used, and where governance boundaries sit. It’s schema-aware, policy-aware, and sits outside your core estate.
From what we’re seeing, this is how most enterprises will roll out “AI infrastructure”: with lightweight, controlled MCP servers that expose just enough surface area to test real use cases - all without rewriting anything. Think of it like a safe staging ground for machine-first interaction.
That has real implications to the AI infrastructure stack.
- Monitoring shifts outward — we’re now being asked to monitor not just the APIs, but the MCP server and its tooling interface.
- Budgets are forming — MCP tools will be scoped, defined, and funded as standalone line items.
- API sprawl gets messier — a proxy on a proxy on a legacy endpoint isn’t neat, but it’s what works today.
And most importantly: observability needs to shift left of the LLM.
We’re already testing this with customers. Monitoring MCP servers, validating tool interfaces, and mapping performance across the full flow: from the client, to the MCP, to the target API. If that sounds messy, that’s because it is. But it’s also where we see the next 12–18 months of enterprise AI adoption going.
There's no AI without APIs. And now, there's no enterprise AI without context.
Get the White Paper
Enterprise API Readiness in the Era of Agentic AI offers a comprehensive playbook to:
- Recognize and mitigate the risks agentic clients introduce
- Build future-facing infrastructure that supports scalable AI use
- Ensure your APIs remain secure, governable, and easy to consume by autonomous systems
This is a transformation in how digital systems interact—and your APIs are at the center.
So how do you prepare your APIs for this new framework? Download our white paper on Enterprise API Readiness in the Era of Agentic AI now!