MC
Database

MCP Server OpenLink ODBC Driver for PyODBC

Connect MCP server to generic DBMS via the OpenLink ODBC driver for PyODBC to enable secure, high-performance database access.

View on GitHub
Quick Install
npx -y @OpenLinkSoftware/mcp-pyodbc-server

Overview

The MCP Server OpenLink ODBC Driver for PyODBC bridges an MCP (Model Context Protocol) server with any DBMS that exposes an ODBC interface. It uses OpenLink’s ODBC drivers together with the PyODBC Python library to provide a generic, performant, and secure way for MCP-based agents and tools to run SQL queries against relational databases.

This integration is useful when you want model-driven tooling (LLM agents, data assistants, or automated workflows) to query real databases without building custom adapters for each vendor. By relying on an ODBC driver, the server supports many backends (Oracle, SQL Server, PostgreSQL, MySQL, commercial gateways) while centralizing connection, authentication, and security configuration.

Features

  • Generic DBMS support via OpenLink ODBC drivers and PyODBC
  • Secure connection options (DSN/TLS, OS authentication, encrypted credentials)
  • Connection pooling and efficient cursor reuse for high throughput
  • Configurable query execution limits (timeout, row limits) to protect production systems
  • Read-only or restricted-mode support for safe agent access
  • Simple HTTP/JSON API for MCP tool integration
  • Example configs and Docker-friendly setup

Installation / Configuration

Prerequisites:

  • OpenLink ODBC driver installed for your target DBMS (vendor packages)
  • Python 3.8+
  • PyODBC and other Python dependencies

Clone and install the project (example):

git clone https://github.com/OpenLinkSoftware/mcp-pyodbc-server.git
cd mcp-pyodbc-server
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

ODBC DSN / connection setup:

  • Create an ODBC Data Source Name (DSN) using your platform’s ODBC manager or configure a DSN-less connection string.
  • Example (DSN-less) connection string format for PyODBC:
conn_str = (
    "DRIVER={OpenLink ODBC Driver};"
    "SERVER=db.example.com;PORT=1234;"
    "DATABASE=mydb;"
    "UID=myuser;PWD=mypassword;"
    "Encrypt=yes;TrustServerCertificate=no;"
)

Server configuration (yaml JSON example):

  • Create a config file (config.yml) that the MCP server will read. Typical entries include DSN/connection string, pooling options, security, and execution limits.
odbc:
  dsn: "MyOpenLinkDSN"          # DSN name or leave empty to use connection_string
  connection_string: ""         # Fallback DSN-less string if DSN is empty
  max_connections: 10
  query_timeout_seconds: 30
  readonly: true                # Enforce read-only sessions if supported

server:
  host: 0.0.0.0
  port: 8080
  api_key: "change_this_to_secure_key"

Start the server (example):

  • The repository includes a runnable entry point. Replace the module/command name as appropriate for your deployment.
# Example: run via python module
python -m mcp_pyodbc_server --config config.yml

# Or build and run via Docker
docker build -t mcp-pyodbc-server .
docker run -p 8080:8080 \
  -v $(pwd)/config.yml:/app/config.yml \
  -e ODBCINI=/etc/odbc.ini \
  mcp-pyodbc-server

Environment variables

VariablePurpose
ODBCINIPath to odbc.ini for DSN definitions
ODBCSYSINIDirectory containing driver manager config
MCP_CONFIGAlternative path to server config file

Available Resources

  • GitHub repository: https://github.com/OpenLinkSoftware/mcp-pyodbc-server
  • OpenLink ODBC Drivers: vendor download pages (install driver for your DB)
  • PyODBC documentation: https://github.com/mkleehammer/pyodbc
  • ODBC driver manager docs (unixODBC / iODBC) for DSN setup

Use Cases

  1. LLM Data Assistant (read-only analytics)
  • Deploy the MCP server in front of a production data warehouse with the server configured to read-only mode and query timeouts. LLM agents can generate safe, audited SQL to answer business questions without direct DB credentials.
  1. Multi-DB Gateway for Applications
  • Use the MCP server as a single API endpoint that routes queries to multiple backends (via different DSNs). Applications and automation tools call the MCP endpoint and avoid embedding different DB drivers or connection logic.
  1. Secure Remote Access for Third Parties
  • Provide partners with an MCP tool API key and tightly-scoped query limits. Partners can query aggregated views via OpenLink drivers without receiving direct DB access or credentials.
  1. Schema Exploration and Autocomplete
  • Combine MCP server results with tools that generate schema summaries for code-completion or agent planning. The server returns column metadata and sample rows using native ODBC metadata calls.

Example API call (JSON to server endpoint):

curl -X POST http://localhost:8080/query \
  -H "Authorization: Bearer <API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{"sql":"SELECT id, name FROM customers WHERE created_at > ? LIMIT 10", "params":["2023-01-01"]}'

Response (typical):

{
  "columns": ["id", "name"],
  "rows": [[1, "Acme"], [2, "Beta Corp"]],
  "execution_time_ms": 42
}

Notes & Best Practices

  • Always use secure credentials handling (secret managers or OS-level protections) instead
Tags:database