Price transparency data has transformed what’s possible in healthcare analytics—but only for those with the technical expertise to access it. The next frontier is making that data conversational. At Serif Health, we’re exploring how new AI frameworks like the Model Context Protocol (MCP) can give large language models the tools they need to understand and interact with healthcare cost data directly.

Serif Health is always looking for ways to make Transparency in Coverage (TiC) data more accessible and actionable. Our latest exploration focuses on how the Model Context Protocol (MCP) could help Large Language Models (LLMs) more seamlessly access our core healthcare cost datasets–and in turn, make it easier for users to query complex data using natural language.

What is MCP?

For a while now, developers of popular Large Language Models (LLMs), and the applications that leverage them such as Claude and ChatGPT, have been looking for ways to bring real-world context into AI interactions. We’ve seen:

• Retrieval-Augmented Generation (RAG) for internal document search
• Web search integration for public information access
• And now, the Model Context Protocol (MCP) for structured data

MCP, an open protocol from Anthropic, allows LLMs to securely interact with protected external sources of structured data (like APIs). It defines a common language for how clients (LLMs) and servers (data systems) can discover and use tools, creating a standardized bridge between them.

Before MCP, developers had to build custom connectors and prompt templates for every API—work that often needed to be redone for each model or vendor. MCP removes that friction by defining a universal format for tool discovery, input/output schemas, and communication.

In short: MCP gives LLMs “hands” to reach out and interact with real data.

How MCP Simplifies API Access

Before MCP, developers would have to write custom “glue code” or connectors for each API. MCP provides a universal format and communication layer that both the LLM and external systems understand.

To understand the power of MCP, it helps to compare a traditional API integration with one powered by MCP:

By connecting an LLM to an MCP client, the model automatically discovers the tools that give it access to our API endpoints—including all the necessary context to select the right endpoint, structure a query, and interpret the results.

Why It Matters for Healthcare Data

For us, the seamlessness of MCP unlocks an especially powerful use case: natural language querying of price transparency data. 

Traditionally, generating insights from large healthcare datasets requires complex filters and parameters. With MCP, we can give an LLM a well-defined schema and let it intelligently guide the parameter selection process. This reduces friction both for users who need quick, data-driven answers, as well for the developers integrating price data into their applications.

This potential motivated us to research, prototype, and ultimately build a production-grade MCP server that connects to our Find Care API, which surfaces provider-level insights by payer, network, procedure, and location. And we’re already planning to expand MCP access to our Neuron API for payer and hospital data in the near future.

See It in Action

To see what this looks like in practice, check out the demo video below — where Claude uses our Find Care tools (find_care_networks, find_care_procedures, and find_care_rates) to power an intelligent care navigation experience.

Try It Yourself

Our Find Care MCP server is now available in public preview. While we’re excited about the potential for natural language search, we’re even more excited to see what you build with it.

If you’d like to try the Serif Health MCP server or explore our broader API suite for price transparency, get in touch at hello@serifhealth.com.

This post is part one of a two-part series:

• Part 1: The Promise of MCP — What MCP is and why it matters for price transparency.
• Part 2: (Stay tuned!)Building MCP in Production — How we took the protocol from prototype to a scalable, production-ready implementation at Serif.