FHIR MCP Server

FHIR MCP Server Overview

What is FHIR MCP Server about?

FHIR MCP Server provides a bridge between standard FHIR APIs and the Model Context Protocol, allowing developers and AI tools to query, read, create, update, and delete clinical resources using MCP transports (stdio, SSE, or streamable HTTP). It abstracts authentication, response filtering, and transport details so that downstream tools can focus on the clinical data they need.

How to use FHIR MCP Server?

Installation

PyPI: uvx fhir-mcp-server
Source: Clone the repo, create a virtual environment, install dependencies with uv pip sync requirements.txt, then run uv run fhir-mcp-server.
Docker: Pull wso2/fhir-mcp-server:latest or build locally, then start with docker run --env-file .env -p 8000:8000 fhir-mcp-server.
Docker Compose: Use the provided docker-compose.yml to spin up both a HAPI FHIR server and the MCP server together.

Configuration

Set FHIR_SERVER_BASE_URL (required). Optional variables include FHIR_SERVER_CLIENT_ID, FHIR_SERVER_CLIENT_SECRET, FHIR_SERVER_SCOPES, FHIR_SERVER_DISABLE_AUTHORIZATION, FHIR_MCP_HOST, FHIR_MCP_PORT, etc.
Choose a transport via the --transport flag (stdio, sse, streamable-http). Default is streamable-http.
Adjust logging with --log-level.

Running

uv run fhir-mcp-server --transport streamable-http --log-level DEBUG

Clients (VS Code, Claude Desktop, MCP Inspector) can connect using the appropriate URL (http://localhost:8000/mcp for HTTP, /sse for SSE, or stdio command).

Key Features

MCP‑compatible transports: stdio, Server‑Sent Events, and streamable HTTP.
SMART‑on‑FHIR based OAuth2 support; can be disabled for open servers.
Response filtering with FHIRPath expressions to trim payloads.
Ready‑made tool definitions (search, read, create, update, delete, get_capabilities, get_user).
Simple integration with popular MCP clients (VS Code, Claude Desktop, MCP Inspector).
Docker and Docker‑Compose setups for instant development environments.

Use Cases

AI assistants retrieving patient demographics or lab results without exposing full FHIR bundles.
Clinical research pipelines that need to batch‑search observations while minimizing network traffic.
LLM‑driven decision support where the model calls search or read tools to obtain specific clinical facts.
Rapid prototyping of healthcare integrations where developers can point the server at any existing FHIR endpoint (public HAPI server, Epic sandbox, etc.).
Secure enterprise deployments leveraging SMART‑on‑FHIR OAuth2 to enforce patient‑level access controls.

FAQ

Q: Which FHIR servers are supported? A: Any server that implements the FHIR REST API, including public HAPI servers, Epic sandbox, or self‑hosted installations.

Q: How do I disable authentication? A: Set the environment variable FHIR_SERVER_DISABLE_AUTHORIZATION=True before starting the server.

Q: What transport should I choose? A: Use streamable-http for typical HTTP client libraries, sse for event‑driven clients, or stdio when embedding the server in a CLI pipeline.

Q: Can I run the server without Docker? A: Yes, the PyPI package runs directly with Python 3.8+ and uvx or a virtual environment.

Q: How are response sizes reduced? A: Provide response_filter_fhirpaths in tool calls; the server applies the FHIRPath expressions and returns only the selected elements.

Q: Where are the tool definitions documented? A: The README lists each tool (search, read, create, update, delete, get_capabilities, get_user) with required and optional parameters.

FHIR MCP Server's README

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

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
Response Filtering using FHIRPath: Filter resources and bundles returned by read and search operations using custom FHIRPath expressions to retrieve only the fields needed for the task, reducing payload sizes.
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, or by cloning this repository.

Installing using PyPI Package

Configure Environment Variables:

To run the server, you must set FHIR_SERVER_BASE_URL.
- To enable authorization: Set FHIR_SERVER_BASE_URL, FHIR_SERVER_CLIENT_ID, FHIR_SERVER_CLIENT_SECRET, and FHIR_SERVER_SCOPES. Authorization is enabled by default.
- To disable authorization: Set FHIR_SERVER_DISABLE_AUTHORIZATION to True.
By default, the MCP server runs on http://localhost:8000, and you can customize the host and port using FHIR_MCP_HOST and FHIR_MCP_PORT.

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"
```
Install the PyPI package and run the server
```
uvx fhir-mcp-server
```

Installing from Source

Clone the repository:

git clone <repository_url>
cd <repository_directory>

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

Configure Environment Variables: Copy the example file and customize if needed:
```
cp .env.example .env
```
Run the server:
```
uv run fhir-mcp-server
```

Installing using Docker

Running the MCP Server with Docker

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

Note on Authorization: When running the MCP server locally via Docker or Docker Compose, authorization should be disabled by setting the environment variable, FHIR_SERVER_DISABLE_AUTHORIZATION=True . This would be fixed in the future releases.

Build the Docker Image or pull the docker image from the container registry:
- Build from source:
```
docker build -t fhir-mcp-server .
```
- Pull from GitHub Container Registry:
```
docker pull wso2/fhir-mcp-server:latest
```
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. See the Configuration section for details on available environment variables.
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.

Using Docker Compose with HAPI FHIR Server

For a quick setup that includes both the FHIR MCP server and a HAPI FHIR server (with PostgreSQL), use the provided docker-compose.yml. This sets up an instant development environment for testing FHIR operations.

Prerequisites:
- Docker and Docker Compose installed.
Run the Stack:
```
docker-compose up -d
```
This command will:
- Start a PostgreSQL database container.
- Launch the HAPI FHIR server (connected to PostgreSQL) listening on http://localhost:8080.
- Build and run the FHIR MCP server container listening on http://localhost:8000, with FHIR_SERVER_BASE_URL set to http://hapi-r4-postgresql:8080/fhir.
Access the Services:
- FHIR MCP Server: http://localhost:8000
- HAPI FHIR Server: http://localhost:8080
- To stop run docker-compose down.
Configure Additional Environment Variables:

If you need to customize OAuth or other settings, adjust the env variables in the docker-compose.yml. The compose file sets basic configuration; refer to the Configuration section for full options.

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 MCP configuration file in VS Code (> V1.104). You can do this by pressing Ctrl + Shift + P and typing MCP: Open User Configuration.

"servers": {
    "fhir": {
        "type": "http",
        "url": "http://localhost:8000/mcp",
    }
}

"servers": {
    "fhir": {
        "command": "uv",
        "args": [
            "--directory",
            "/path/to/fhir-mcp-server",
            "run",
            "fhir-mcp-server",
            "--transport",
            "stdio"
        ],
        "env": {
            "FHIR_SERVER_ACCESS_TOKEN": "Your FHIR Access Token"
        }
    }
}

"servers": {
    "fhir": {
        "type": "sse",
        "url": "http://localhost:8000/sse",
    }
}

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:

{
    "mcpServers": {
        "fhir": {
            "command": "npx",
            "args": [
                "-y",
                "mcp-remote",
                "http://localhost:8000/mcp"
            ]
        }
    }
}

{
    "mcpServers": {
        "fhir": {
            "command": "uv",
            "args": [
                "--directory",
                "/path/to/fhir-mcp-server",
                "run",
                "fhir-mcp-server",
                "--transport",
                "stdio"
            ],
            "env": {
                "FHIR_SERVER_ACCESS_TOKEN": "Your FHIR Access Token"
            }
        }
    }
}

{
    "mcpServers": {
        "fhir": {
            "command": "npx",
            "args": [
                "-y",
                "mcp-remote",
                "http://localhost:8000/sse"
            ]
        }
    }
}

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
--help
- Description: Displays a help message with available server options and exits.
- Usage: Automatically provided by the command-line interface.

Sample Usages:

uv run fhir-mcp-server --transport streamable-http --log-level DEBUG
uv run fhir-mcp-server --help

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_DISABLE_AUTHORIZATION: If set to True, disables authorization checks on the MCP server, allowing connections to publicly accessible FHIR servers.
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"}).
- response_filter_fhirpaths: (Optional) An array of FHIRPath expressions (e.g., ["Patient.name", "Patient.birthDate", "Bundle.link.where(relation='next').url"]) to apply to the resources in the response bundle.
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").
- response_filter_fhirpaths: (Optional) An array of FHIRPath expressions (e.g., ["Patient.name", "Observation.valueQuantity"]) to filter the returned single resource (or entries when using custom operations like $everything).
create: Executes a FHIR "create" interaction to persist a new resource of the specified type.
- type: The FHIR resource type name (e.g., "Device", "CarePlan", "Goal").
- payload: A JSON object representing the full FHIR resource body to be created.
- searchParam: A mapping of FHIR search parameter names to their desired values (e.g., {"address-city":"Boston"}).
- operation: The name of a custom FHIR operation or extended query defined for the resource (e.g., "$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:

# Install project in development mode with test dependencies
pip install -e '.[test]'

# Or install from requirements file
pip install -r requirements-dev.txt

Using uv:

# Install development dependencies
uv sync --dev

Running Tests

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

# Simple test runner
python run_tests.py

# Or direct pytest usage
PYTHONPATH=src python -m pytest tests/ -v --cov=src/fhir_mcp_server

Using pytest:

pytest tests/

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.

FHIR MCP Server Reviews

Share Your Experience

Login Required

Please log in to share your review and rating for this MCP.