BurriRitvik-EMIDS/MCP_Server

FEP-MCP Server

A FastMCP server for healthcare data analysis with member search, claims analysis, and database operations.

Quick Start

python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
python server.py [for running MCP Server]
python client.py [for testing MCP Server]

Environment Setup

Create a .env file in the root directory with the following variables:

# Required for database access
SQL_SERVER_NAME=your-server.database.windows.net
SQL_DATABASE_NAME=your-database
SQL_DB_USERNAME=your-username
SQL_DB_PASSWORD=your-password

# Required for vector store (PGVector)
POSTGRES_USER=your_postgres_username
POSTGRES_PASSWORD=your_postgres_password
POSTGRES_HOST=your_postgres_host  # e.g., localhost or server-address
POSTGRES_DB=your_database_name

# Optional: Override default timeout (in seconds)
MCP_SERVER_REQUEST_TIMEOUT=5

# Optional: SQL Driver version (defaults to "ODBC Driver 18 for SQL Server")
# SQL_DRIVER=ODBC Driver 18 for SQL Server

Notes on Environment Variables

• The SQL_* variables are used for the main application database
• The POSTGRES_* variables are specifically for the vector store functionality
• Make sure the PostgreSQL server has the vector extension installed
• The database user must have appropriate permissions to create and modify tables

Data Sources

The server works with the following data sources:

• Claims data from CSV (local or remote)
• Member data from JSON API
• Clinical notes from database
• Vector store for semantic search

Available Tools

Member Management

• find_member

  • Description: Find full member record by member_id.
  • Inputs:
    • member_id (string)

• search_member_key

  • Description: Search for a partial key inside a member's record.
  • Inputs:
    • member_id (string)
    • search_key (string)

• get_basic_info

  • Description: Return basic member info (for customer service) given a member_id.
  • Inputs:
    • member_id (string)

Data Analysis

• run_claims_analysis

  • Description: Execute Python code against claims CSV file.
  • Inputs:
    • python_code (string)

• run_graph_generation

  • Description: Execute Python code for graph generation with access to both claims and clinical notes data. Execute Python code that generates a matplotlib plot, and return the image. The code must generate a visual output.
  • Inputs:
    • python_code (string)

Database Operations

• get_db_schema

  • Description: Retrieve database schema (tables & columns).

• run_read_query

  • Description: Execute a read-only query against the database.
  • Inputs:
    • query (string)
    • params (object, optional)

• run_write_query

  • Description: Execute a write (INSERT/UPDATE/DELETE) query against the database.
  • Inputs:
    • query (string)
    • params (object, optional)

• get_clinical_notes

  • Description: Fetch clinical notes for a member from table clinical_report by member_id.
  • Inputs:
    • member_id (string)

Vector Store

• retrieve_from_vectorstore
  • Description: Semantic retrieve top-k chunks from the configured PGVector store. Returns a list of objects with 'page_content' and optional 'metadata'.
  • Inputs:
    • query (string)
    • k (integer, default: 4)

Requirements

• Python 3.12+
• Windows (for SQL Server drivers)
• ODBC Driver 17 or 18 for SQL Server
• Optional: Azure SQL Database access

Dependencies

Core Dependencies

• fastmcp
• langchain_experimental
• typer
• pyodbc
• uv
• requests
• python-dotenv
• pandas
• matplotlib
• numpy

Vector Store Dependencies

• langchain-community
• langchain-postgres
• sentence-transformers
• psycopg2-binary

Notes

• The server attempts to use a local CSV file for claims data if available, otherwise falls back to a remote URL
• Matplotlib is configured to use 'Agg' backend for non-interactive use
• Database operations require proper environment variables to be set