Model Context Protocol (MCP) Server for Fast Healthcare Interoperability Resources (FHIR) APIs

Table of Contents

• Overview

• Demo

  • Demo with HAPI FHIR server

  • Demo with EPIC Sandbox

• Core Features

• Prerequisites

• Installation

  • Installing using PyPI Package

  • Installing from Source

  • Installing using Docker

• Integration with MCP Clients

  • VS Code

  • Claude Desktop

  • MCP Inspector

• Configuration

  • CLI Options

  • Environment Variables

• Tools

• Development & Testing

  • Installing Development Dependencies

  • Running Tests

Overview

The FHIR MCP Server is a Model Context Protocol (MCP) server that provides seamless integration with FHIR APIs. Designed for developers, integrators, and healthcare innovators, this server acts as a bridge between modern AI/LLM tools and healthcare data, making it easy to search, retrieve, and analyze clinical information.

Demo

Demo with HAPI FHIR server

This video showcases the MCP server's functionality when connected to a public HAPI FHIR server. This example showcases direct interaction with an open FHIR server that does not require an authorization flow.

https://github.com/user-attachments/assets/cc6ac87e-8329-4da4-a090-2d76564a3abf

Demo with EPIC Sandbox

This video showcases the MCP server's capabilities within the Epic EHR ecosystem. It demonstrates the complete OAuth 2.0 Authorization Code Grant flow.

https://github.com/user-attachments/assets/96b433f1-3e53-4564-8466-65ab48d521de

Core Features

• MCP-compatible transport: Serves FHIR via stdio, SSE, or streamable HTTP

• SMART-on-FHIR based authentication support: Securely authenticate with FHIR servers and clients

• Tool integration: Integratable with any MCP client such as VS Code, Claude Desktop, and MCP Inspector

Prerequisites

• Python 3.8+

• uv (for dependency management)

• An accessible FHIR API server.

Installation

You can use the FHIR MCP Server by installing our Python package, by cloning this repository or by running as a docker container.

Installing using PyPI Package

1. Configure Environment Variables:

   To run the server, you must set FHIR_SERVER_BASE_URL. If you plan to use authorization, you'll also need to configure FHIR_SERVER_CLIENT_ID, FHIR_SERVER_CLIENT_SECRET, and FHIR_SERVER_SCOPES. By default, the MCP server will listen on http://localhost:8000. You can customize the host and port by setting FHIR_MCP_HOST and FHIR_MCP_PORT respectively.

   You can set these by exporting them as environment variables like below or by creating a .env file (referencing .env.example).

   export FHIR_SERVER_BASE_URL="" export FHIR_SERVER_CLIENT_ID="" export FHIR_SERVER_CLIENT_SECRET="" export FHIR_SERVER_SCOPES="" export FHIR_MCP_HOST="localhost" export FHIR_MCP_PORT="8000"

2. Install the PyPI package and run the server

   If you are exposing a FHIR server with no security:

   uvx fhir-mcp-server --disable-auth

   else;

   uvx fhir-mcp-server

Installing from Source

1. Clone the repository:

   git clone <repository_url> cd <repository_directory>

2. Create a virtual environment and install dependencies:

   uv venv source .venv/bin/activate uv pip sync requirements.txt

   Or with pip:

   python -m venv .venv source .venv/bin/activate pip install -r requirements.txt

3. Configure Environment Variables: Copy the example file and customize if needed:

   cp .env.example .env

4. Run the server:

   uv run fhir-mcp-server

Installing using Docker

You can run the MCP server using Docker for a consistent, isolated environment.

1. Build the Docker Image

   docker build -t fhir-mcp-server .

2. Configure Environment Variables

   Copy the example environment file and edit as needed:

   cp .env.example .env # Edit .env to set your FHIR server, client credentials, etc.

   Alternatively, you can pass environment variables directly with -e flags or use Docker secrets for sensitive values.

3. Run the Container

   docker run --env-file .env -p 8000:8000 fhir-mcp-server

   This will start the server and expose it on port 8000. Adjust the port mapping as needed.

Integration with MCP Clients

The FHIR MCP Server is designed for seamless integration with various MCP clients.

VS Code

Add the following JSON block to your User Settings (JSON) file in VS Code (> V1.101). You can do this by pressing Ctrl + Shift + P and typing Preferences: Open User Settings (JSON).

Claude Desktop

Add the following JSON block to your Claude Desktop settings to connect to your local MCP server.

• Launch the Claude Desktop app, click on the Claude menu in the top bar, and select "Settings…".

• In the Settings pane, click “Developer” in the left sidebar. Then click "Edit Config". This will open your configuration file in your file system. If it doesn’t exist yet, Claude will create one automatically at:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

  • Windows: %APPDATA%\Claude\claude_desktop_config.json

• Open the claude_desktop_config.json file in any text editor. Replace its contents with the following JSON block to register the MCP server:

MCP Inspector

Follow these steps to get the MCP Inspector up and running:

• Open a terminal and run the following command:

  npx -y @modelcontextprotocol/inspector

• In the MCP Inspector interface:

• Transport Type: Streamable HTTP

• URL: http://localhost:8000/mcp

• Transport Type: STDIO

• Command: uv

• Arguments: --directory /path/to/fhir-mcp-server run fhir-mcp-server --transport stdio

• Transport Type: SSE

• URL: http://localhost:8000/sse

Make sure your MCP server is already running and listening on the above endpoint.

Once connected, MCP Inspector will allow you to visualize tool invocations, inspect request/response payloads, and debug your tool implementations easily.

Configuration

CLI Options

You can customize the behavior of the MCP server using the following command-line flags:

• --transport

  • Description: Specifies the transport protocol used by the MCP server to communicate with clients.

  • Accepted values: stdio, sse, streamable-http

  • Default: streamable-http

• --log-level

  • Description: Sets the logging verbosity level for the server.

  • Accepted values: DEBUG, INFO, WARN, ERROR (case-insensitive)

  • Default: INFO

• --disable-auth

  • Description: Disables the security of the MCP Server. Allows you to connect with openly available FHIR servers.

  • Type: Flag (no value required)

  • Default: False (authentication enabled)

• --help

  • Description: Displays a help message with available server options and exits.

  • Usage: Automatically provided by the command-line interface.

Sample Usages:

Environment Variables

MCP Server Configurations:

• FHIR_MCP_HOST: The hostname or IP address the MCP server should bind to (e.g., localhost for local-only access, or 0.0.0.0 for all interfaces).

• FHIR_MCP_PORT: The port on which the MCP server will listen for incoming client requests (e.g., 8000).

• FHIR_MCP_SERVER_URL: If set, this value will be used as the server's base URL instead of generating it from host and port. Useful for custom URL configurations or when behind a proxy.

• FHIR_MCP_REQUEST_TIMEOUT: Timeout duration in seconds for requests from the MCP server to the FHIR server (default: 30).

MCP Server OAuth2 with FHIR server Configuration (MCP Client ↔ MCP Server): These variables configure the MCP client's secure connection to the MCP server, using the OAuth2 authorization code grant flow with a FHIR server.

• FHIR_SERVER_CLIENT_ID: The OAuth2 client ID used to authorize MCP clients with the FHIR server.

• FHIR_SERVER_CLIENT_SECRET: The client secret corresponding to the FHIR client ID. Used during token exchange.

• FHIR_SERVER_BASE_URL: The base URL of the FHIR server (e.g., https://hapi.fhir.org/baseR4). This is used to generate tool URIs and to route FHIR requests.

• FHIR_SERVER_SCOPES: A space-separated list of OAuth2 scopes to request from the FHIR authorization server (e.g., user/Patient.read user/Observation.read). Add fhirUser openid to enable retrieval of user context for the get_user tool. If these two scopes are not configured, the get_user tool returns an empty result because the ID token lacks the user's FHIR resource reference.

• FHIR_SERVER_ACCESS_TOKEN: The access token to use for authenticating requests to the FHIR server. If this variable is set, the server will bypass the OAuth2 authorization flow and use this token directly for all requests.

Tools

• get_capabilities: Retrieves metadata about a specified FHIR resource type, including its supported search parameters and custom operations.

  • type: The FHIR resource type name (e.g., "Patient", "Observation", "Encounter")

• search: Executes a standard FHIR search interaction on a given resource type, returning a bundle or list of matching resources.

  • type: The FHIR resource type name (e.g., "MedicationRequest", "Condition", "Procedure").

  • searchParam: A mapping of FHIR search parameter names to their desired values (e.g., {"family":"Simpson","birthdate":"1956-05-12"}).

• read: Performs a FHIR "read" interaction to retrieve a single resource instance by its type and resource ID, optionally refining the response with search parameters or custom operations.

  • type: The FHIR resource type name (e.g., "DiagnosticReport", "AllergyIntolerance", "Immunization").

  • id: The logical ID of a specific FHIR resource instance.

  • searchParam: A mapping of FHIR search parameter names to their desired values (e.g., {"device-name":"glucometer"}).

  • operation: The name of a custom FHIR operation or extended query defined for the resource (e.g., "$everything").

• create: Executes a FHIR "create" interaction to persist a new resource of the specified type.

  • type: The FHIR resource type name (e.g., "Device", "CarePlan", "Goal").

  • payload: A JSON object representing the full FHIR resource body to be created.

  • searchParam: A mapping of FHIR search parameter names to their desired values (e.g., {"address-city":"Boston"}).

  • operation: The name of a custom FHIR operation or extended query defined for the resource (e.g., "$evaluate").

• update: Performs a FHIR "update" interaction by replacing an existing resource instance's content with the provided payload.

  • type: The FHIR resource type name (e.g., "Location", "Organization", "Coverage").

  • id: The logical ID of a specific FHIR resource instance.

  • payload: The complete JSON representation of the FHIR resource, containing all required elements and any optional data.

  • searchParam: A mapping of FHIR search parameter names to their desired values (e.g., {"patient":"Patient/54321","relationship":"father"}).

  • operation: The name of a custom FHIR operation or extended query defined for the resource (e.g., "$lastn").

• delete: Execute a FHIR "delete" interaction on a specific resource instance.

  • type: The FHIR resource type name (e.g., "ServiceRequest", "Appointment", "HealthcareService").

  • id: The logical ID of a specific FHIR resource instance.

  • searchParam: A mapping of FHIR search parameter names to their desired values (e.g., {"category":"laboratory","issued:"2025-05-01"}).

  • operation: The name of a custom FHIR operation or extended query defined for the resource (e.g., "$expand").

• get_user: Retrieves the currently authenticated user's FHIR resource (for example the linked Patient resource) and returns a concise profile containing available demographic fields such as id, name, and birthDate.

Development & Testing

Installing Development Dependencies

To run tests and contribute to development, install the test dependencies:

Using pip:

Using uv:

Running Tests

The project includes a comprehensive test suite covering all major functionality:

Using pytest:

This will discover and run all tests in the tests/ directory.

Test Features:

• 100+ tests with comprehensive coverage

• Full async/await support using pytest-asyncio

• Complete mocking of HTTP requests and external dependencies

• Coverage reporting with terminal and HTML output

• Fast execution with no real network calls

The test suite includes:

• Unit tests: Core functionality testing

• Integration tests: Component interaction validation

• Edge case coverage: Error handling and validation scenarios

• Mocked OAuth flows: Realistic authentication testing

Coverage reports are generated in htmlcov/index.html for detailed analysis.