%20(1).jpg)
For decades, companies have poured countless product and engineering hours into a problem that sounds simple but rarely is: how do you make sure your employees actually engage with the tools you've built for them? The answer has largely followed the same formula: design an intuitive platform, invest in a great front-end experience, and trust that a genuinely valuable product will earn its place in people's daily workflows. For a long time, the UI was the destination. But the goalposts are moving. A quiet shift is now underway, powered by a new technology that most people outside the AI world haven't come across yet: Model Context Protocol (MCP).
MCP is an open standard that allows AI assistants like Claude or ChatGPT or Gemini to connect directly to external tools, data sources, and services. Think of it as a universal plug that lets an AI assistant reach out and actually do things in the world: query a database, retrieve a document, call an API, or interact with a third-party platform.
If AI assistants are the new platform through which people get things done, then MCP servers are the infrastructure that makes those interactions meaningful. They're the bridge between a user's intent and a company's capability.
Before MCP, delivering value to end users meant owning their attention. You needed them to open your app, navigate your UI, and remember that your product existed in the first place. Even well-designed platforms struggle with this. At the end of the day, a product is only as useful as its adoption, and adoption is hard to sustain when users are already juggling a dozen other platforms.
The result? Valuable features go undiscovered. Users default to whatever is easiest. And companies watch engagement metrics plateau despite continued investment.
MCP changes the equation by enabling companies to deliver their solutions inside the AI tools users are already using. Rather than pulling users toward a separate platform, your functionality meets them where they are and lets them engage with a solution in ways that are comfortable and familiar.
We like to think of this new avenue of distribution not as a single door but a custom door for each user. The more you use your AI agent, the more it becomes familiar with how you like to consume data. Some people visualize things in different ways. The challenge with traditional user experiences is that you are delivering one experience that needs to meet the needs of a broad population. MCP flips this on its head: Now, we can deliver the data to the AI assistant and it can use what it knows about the user to display that in the most meaningful way. Additionally, users are empowered to prompt and ask for the data to be displayed in different ways.
We recently shipped a feature that puts this into practice in a concrete and high-impact way: a provider search tool, delivered via our MCP server, that allows users to find the right healthcare provider without ever leaving their AI assistant.
Finding the right doctor has always been more painful than it should be. Employees typically have to log into a benefits portal, cross-reference their insurance network, search by location, filter by specialty, and then try to assess quality. This is often done across multiple tabs and systems. It's friction at exactly the moment someone needs help.
Our MCP server eliminates that friction. Through a natural conversation with an AI assistant, a user can now ask something as simple as: "Can you find me an in-network dermatologist near downtown Chicago with good reviews?" and get a precise, personalized answer in seconds. They can prompt the AI assistant to display the data in maps, cards, or whatever format allows them to visualize the data in the best way.
Shipping an MCP tool isn't just a technical exercise, it requires deliberate thinking about how the AI assistant communicates with the user. We worked closely with the AI layer to ensure that the experience felt natural, accurate, and trustworthy.
That meant a few things in practice. First, we designed the tool's inputs and outputs to align with how people actually ask questions in conversation, not how a database query is structured. The AI needed to be able to interpret a range of phrasings and intents and map them reliably to the right parameters.
Second, we thought carefully about how results are presented. A raw list of provider names isn't useful. We worked to ensure the AI surfaces results with the right context: network status, distance, specialty, and rating all in a format that helps users make decisions quickly.
Finally, we built in guardrails to handle edge cases gracefully: what happens when no in-network providers are found in a given area, or when a user's plan information is incomplete. The goal was an experience that felt helpful even when the answer wasn't straightforward.
The provider search tool is one example of a broader principle: that the most valuable features are the ones people actually use. By delivering this capability through an AI assistant via MCP, we removed the biggest barrier to engagement: the requirement that users seek out the tool in the first place.
This is what MCP makes possible at scale. Companies no longer have to choose between building great functionality and ensuring it gets used. The AI assistant becomes the distribution layer, and MCP is what connects your product to it.
The door between enterprise solutions and the people they're meant to serve has always been there. Building an MCP server is one way you can walk through it.
