Hospital Assistant MCP Server
noopur-zambare/hospital-assistant-MCP-server
MCP dual-server para lookup clínico, ICD-10 e pesquisa médica com RAG
Ver no GitHubSistema dual de servidores MCP com agente ReAct em LangGraph que integra APIs de conhecimento médico externo com banco de dados local de pacientes. Oferece lookup clínico, codificação ICD-10, busca de ensaios clínicos e acesso a pesquisas médicas recentes, com interface Streamlit para suporte à decisão clínica.
Instalação
npx convez-ai noopur-zambare/hospital-assistant-MCP-server1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Lookup de códigos ICD-10-CM e tópicos de saúde
- Busca em banco de dados local de pacientes e resultados de laboratório
- Pesquisa de ensaios clínicos ativos por condição
- Recuperação de preprints médicos recentes (medRxiv)
- Suporte à decisão clínica com rastreamento completo de ferramentas
Veredito

Implementação sólida de arquitetura MCP multi-servidor com componentes bem estruturados (RAG, APIs externas, banco local). Prototipagem educacional respeitável, mas requer rigor regulatório e validação clínica antes de uso em produção hospitalar real — adequado para pesquisa e prova de conceito.
README
A tradução para o português deste README ainda está na fila.
Hospital Assistant MCP Server
A dual MCP-server system built with a LangGraph ReAct agent, combining external medical knowledge APIs with a local patient database for clinical lookups, ICD-10 coding, and research-backed responses.
Demo
https://github.com/user-attachments/assets/7ac75e7d-74b0-4f79-aab3-7788b9911056
System Overview
The project runs four components:
- External MCP Server - public medical knowledge APIs
- Internal MCP Server - local patient database
- RAG Layer (FAISS) - unstructured clinical notes retrieval
- Streamlit App - chat UI with an OpenAI agent that calls tools across both servers
External MCP Server
Exposes public medical APIs as MCP tools.
Integrated APIs
| API | Purpose |
|---|---|
| Health.gov MyHealthfinder | Consumer health topics |
| NLM Clinical Tables (ICD-10-CM) | Diagnosis code lookup |
| ClinicalTrials.gov v2 | Active and completed trials |
| medRxiv | Recent medical research preprints |
Tools
-
health_topics
Search Health.gov MyHealthfinder for consumer health information by keyword.
-
icd10_lookup
Look up ICD-10-CM codes by name or keyword via the NLM Clinical Tables API.
-
clinical_trials
Search ClinicalTrials.gov v2 for studies by condition.
-
medrxiv_search
Pull recent medRxiv preprints matching a query (180-day window).
Internal MCP Server
Serves a local patient database and search over patient notes.
Tools
-
get_patient
Fetch a patient's record by ID (e.g. P003).
-
get_lab_results
Retrieve a patient's lab test results.
-
get_risk_score
Get a patient's calculated clinical risk score.
-
list_patients
List all patients in the internal database.
Features
- Multi-server MCP client - talks to both external (8000) and internal (8001) servers in one agent.
- Tool trace per turn - every assistant reply has an expandable panel showing exactly which tools were called, with what arguments, and what they returned.
- Structured tool outputs - every response is grounded in tool results, fully visible in the trace.
- Decision support, not diagnosis - designed to help clinicians look up information, not to replace clinical judgement.
- Settings panel - switch OpenAI model, adjust temperature, and edit the system prompt live without restarting.
Architecture
hospital-assistant-MCP-server/
│
├── servers/
│ ├── external_server.py
│ ├── internal_server.py
│ └── rag_tool.py
├── app.py
├── .env
├── hospital.db
└── README.md
Running
1. Install dependencies
python -m venv mcp-env
source mcp-env/bin/activate
pip install streamlit langchain-mcp-adapters langgraph langchain-openai \
python-dotenv httpx mcp
2. Set environment variables
Create a .env file:
OPENAI_API_KEY=''
3. Start the MCP servers
In two separate terminals:
python servers/external_server.py
python servers/internal_server.py
4. Launch the UI
streamlit run app.py
Tech Stack
- MCP Servers
- LangGraph
- LangChain
- Streamlit
⚠️ Disclaimer
This project is for research and educational purposes only. It is not a substitute for professional medical advice, diagnosis, or treatment.