AI-FHIR-Bundle

A local, docker-compose based stack that integrates a FHIR server, LLM tooling, web search, and MCP (Model Context Protocol) orchestration—designed for building and testing AI agents that interact with FHIR resources and external web content.

Features

• HAPI FHIR Server (R4) with MCP integration
• PostgreSQL database for persistence
• Open WebUI — chat interface for interacting with local LLMs
• Ollama integration for local LLM inference
• SearXNG — privacy-focused metasearch engine
• Crawl4AI — web crawling and content ingestion service
• fhir-mcp-server - access to HL7 Terminolgy server
• MCPO — MCP Orchestrator/Proxy to manage MCP connections

Prerequisites

• Docker and Docker Compose
• Ollama (if you want local LLM inference)
  • Install: see https://docs.ollama.com/quickstart or, on macOS with Homebrew, brew install ollama
  • Start: follow Ollama docs or run ollama serve if needed
  • Pull a model (example):

To have the best experience you need a GPU with enough RAM to completely load your model of choice. Our tests and default configuration use the qwen3-coder-next model that uses 52GB of GPU RAM!

ollama pull qwen3-coder-next:latest

Configuration files

Place the following files in the project root (examples may be included in the repo):

• searXNG_settings.yml — SearXNG configuration
• mcpo_config.json — MCP orchestrator configuration for connecting to Crawl4AI and other MCP servers

Quick start

1. Copy the environment file and edit values if needed:

cp .env.example .env
# Edit .env if Ollama or other services run on non-default hosts/ports

2. Start the full stack (pulls latest images and removes orphaned containers):

docker compose up --pull always --remove-orphans

This will pull the required images, remove orphaned containers, and start all services defined in docker-compose.yml.

Service endpoints

• Open WebUI: http://localhost:3000
• SearXNG: http://localhost:8080
• HAPI FHIR Server (R4): http://localhost:8081
• Crawl4AI: http://localhost:11235
• fhir-mcp-server: http://localhost:8001
• MCPO: http://localhost:8000
• PostgreSQL: localhost:5432 (default: postgres / password)

Using Open WebUI

1. Open http://localhost:3000
2. In the UI: User → Admin Panel → Settings → Models
   • Select the model you pulled in Ollama and add model_prompt.txt as the model prompt
3. Under Tools, enable the integrations: crawl4ai, HAPI FHIR, terminology-server
4. Save and update settings

Notes:

• Only enable Web Search when you need live web queries — it triggers searches for every requests and can slow interactions.
• Use Crawl4AI when you need deep search as scraping and ingestion takes a long time.

Tips & Troubleshooting

• Prevent the LLM from accidentally trying to change the terminology server as you probably not intending to and the provided server is read only anyways: in Open WebUI go to User → Admin Panel → Settings → External tools → terminology-server and add the following to the Function Name Filter List (comma-separated):

!tool_create_post, !tool_update_post, !tool_delete_post

• If crawls fail frequently, add tool_crawl_post to the Function Name Filter List.

License

See the LICENSE file for license details.