Agent2Agent OMCP

May 9, 2025 ยท View on GitHub

MCP Server

The MCP (Model-Context-Protocol) server provides a set of tools for interacting with OMOP CDM databases and generating SQL queries using natural language processing. The server is built using FastMCP and integrates with Ollama for LLM capabilities.

Features

SQL Server Tools

  • Execute_SQL_Query: Execute SQL queries against an OMOP database and return results in CSV format
  • Test_Connection: Test if a database connection is valid
  • Get_OMOP_Schema: Get the OMOP CDM schema information for prompting

Validation Server Tools

  • Validate_SQL_Query: Validate SQL queries against OMOP CDM validation rules

Ollama Server Tools

  • Generate_SQL: Generate SQL from natural language using an LLM, incorporating medical concept codes
  • Generate_Explanation: Generate an explanation for an SQL query
  • Generate_Answer: Generate a natural language answer based on query, SQL, and results
  • List_Available_Models: List available LLM models from Ollama

Configuration

The server uses a configuration file (config/config.json) that specifies:

  • Database connection strings
  • Schema directory location
  • Ollama API settings
  • OMOP CDM validation rules and schema files

Medical Concept Integration

The SQL generation tool accepts medical concepts in the following format:

{
    "conditions": [
        {
            "concept_id": 12345,
            "concept_name": "Diabetes",
            "vocabulary_id": "SNOMED"
        }
    ],
    "drugs": [
        {
            "concept_id": 11111,
            "concept_name": "Metformin",
            "vocabulary_id": "RxNorm"
        }
    ],
    "measurements": [
        {
            "concept_id": 22222,
            "concept_name": "Blood Pressure",
            "vocabulary_id": "LOINC"
        }
    ]
}

Requirements

  • Python 3.13 or higher
  • PostgreSQL database with OMOP CDM schema
  • Ollama running locally for LLM capabilities

Dependencies

  • mcp
  • httpx
  • sqlalchemy
  • pydantic
  • pydantic-settings
  • python-multipart
  • sse-starlette

Usage

The MCP server can be used as a standalone service or integrated into other applications. To use it:

  1. Ensure all dependencies are installed:
uv pip install -e .

  1. Configure the database connection and other settings in config/config.json

  2. Start the server:

from src.unified_mcp import mcp
mcp.run(transport="stdio")

Example

# Generate SQL with medical concepts
medical_concepts = {
    "conditions": [
        {"concept_id": 12345, "concept_name": "Diabetes", "vocabulary_id": "SNOMED"}
    ]
}

schema = mcp.tools["Get_OMOP_Schema"]()
sql_query, confidence = await mcp.tools["Generate_SQL"](
    prompt="Find all patients with diabetes",
    medical_concepts=medical_concepts,
    schema=schema
)

# Validate the generated SQL
validation_result = mcp.tools["Validate_SQL_Query"](sql_query)

# Execute the query if valid
if validation_result["is_valid"]:
    results = mcp.tools["Execute_SQL_Query"](sql_query)