Migrate from pyodbc to mssql-python

The mssql-python driver is Microsoft's first-party Python driver for Microsoft SQL. If you prefer a Microsoft-maintained driver option, it offers:

• No external ODBC driver dependency.
• Built-in connection pooling.
• Modern Python 3.10+ support.
• Native Microsoft Entra authentication.

Key differences

Basic migration steps

The following steps cover the key changes to migrate a pyodbc application to mssql-python.

1. Update imports

Replace the pyodbc import with mssql_python:

Before (pyodbc):

import pyodbc

After (mssql-python):

import mssql_python

2. Update connection strings

Remove the DRIVER= keyword and update the authentication method:

Before (pyodbc, requires ODBC driver):

conn = pyodbc.connect(
    "DRIVER={ODBC Driver 18 for SQL Server};"
    "SERVER=localhost;"
    "DATABASE=AdventureWorks2022;"
    "Trusted_Connection=yes;"
)

After (mssql-python, no driver needed, using Microsoft Entra auth):

conn = mssql_python.connect(
    "Server=<server>.database.windows.net;"
    "Database=<database>;"
    "Authentication=ActiveDirectoryDefault;"
    "Encrypt=yes;"
)

3. Keep your queries as-is

The mssql-python driver supports both ? (qmark) and %(name)s (pyformat) parameter styles. Your existing ? queries work without changes:

Before (pyodbc):

cursor.execute("SELECT * FROM Production.Product WHERE ProductID = ? AND Color = ?", (1, "Red"))

After (mssql-python, same query):

cursor.execute("SELECT * FROM Production.Product WHERE ProductID = ? AND Color = ?", (1, "Red"))

4. Keep executemany as-is

Existing executemany calls with tuples and ? markers work without changes:

Before (pyodbc):

cursor.execute("CREATE TABLE #MigrateDemo (ID INT, Name NVARCHAR(50))")
data = [(1, "Alice"), (2, "Bob"), (3, "Carol")]
cursor.executemany("INSERT INTO #MigrateDemo (ID, Name) VALUES (?, ?)", data)

After (mssql-python, same code):

cursor.execute("DROP TABLE IF EXISTS #MigrateDemo")
cursor.execute("CREATE TABLE #MigrateDemo (ID INT, Name NVARCHAR(50))")
data = [(1, "Alice"), (2, "Bob"), (3, "Carol")]
cursor.executemany("INSERT INTO #MigrateDemo (ID, Name) VALUES (?, ?)", data)

Stored procedure migration

The mssql-python driver doesn't implement callproc(). The following sections show how to use EXECUTE instead.

Use EXECUTE for stored procedures

The pyodbc driver supports callproc(), but the mssql-python driver doesn't. Use EXECUTE instead:

Before (pyodbc):

cursor.callproc("dbo.uspGetEmployeeManagers", (5,))
results = cursor.fetchall()

After (mssql-python):

cursor.execute(
    "EXECUTE dbo.uspGetEmployeeManagers @BusinessEntityID = %(id)s",
    {"id": 5}
)
results = cursor.fetchall()
print(f"Got {len(results)} rows")

Output parameters

Use T-SQL variables to capture output values instead of relying on callproc() output parameters:

Before (pyodbc, using callproc):

params = (category_id, pyodbc.SQL_INTEGER)
cursor.callproc("dbo.GetProductCount", params)
count = params[1].value

After (mssql-python, using T-SQL variables):

cursor.execute(
    """
    DECLARE @count INT;
    SELECT @count = COUNT(*) FROM Production.Product
    WHERE ProductSubcategoryID = %(cat_id)s;
    SELECT @count AS ProductCount;
    """,
    {"cat_id": 1}
)
product_count = cursor.fetchval()
print(f"Product count: {product_count}")

Feature-specific migrations

The following sections cover specific pyodbc features and their mssql-python equivalents.

Connection strings

Autocommit

The autocommit behavior is identical in both drivers:

pyodbc:

conn.autocommit = True
pyodbc.connect(connection_string, autocommit=True)

mssql-python:

conn.autocommit = True

Bulk inserts

To speed up large INSERT batches, pyodbc users set fast_executemany = True. The mssql-python driver already optimizes executemany for parameterized batches, so moderate inserts need no special flag. For large data loads, prefer bulkcopy(), which streams rows over the bulk-copy protocol and is much faster than issuing individual INSERT statements. For the full workflow, see Use bulk copy.

pyodbc:

cursor.fast_executemany = True
cursor.executemany(query, data)

After (mssql-python), moderate batches with executemany:

cursor.execute("DROP TABLE IF EXISTS #BulkTarget")
cursor.execute("CREATE TABLE #BulkTarget (ID INT, Name NVARCHAR(50))")
data = [(i, f"Item {i}") for i in range(100)]
cursor.executemany("INSERT INTO #BulkTarget (ID, Name) VALUES (?, ?)", data)
conn.commit()

After (mssql-python), large loads with bulkcopy (preferred):

cursor.execute("IF OBJECT_ID('##BulkTarget') IS NOT NULL DROP TABLE ##BulkTarget")
cursor.execute("CREATE TABLE ##BulkTarget (ID INT, Name NVARCHAR(50))")
conn.commit()  # Commit DDL before bulkcopy
data = [(i, f"Item {i}") for i in range(100)]
result = cursor.bulkcopy("##BulkTarget", data)
print(f"Bulk copied {result['rows_copied']} rows")
cursor.execute("DROP TABLE ##BulkTarget")
conn.commit()

Row factory

The mssql-python driver returns Row objects that support attribute access by default, without requiring a custom row factory:

pyodbc (custom row factory):

def namedtuple_row_factory(cursor):
    from collections import namedtuple
    columns = [col[0] for col in cursor.description]
    Row = namedtuple("Row", columns)
    return Row

mssql-python (attribute access by default):

cursor.execute("SELECT Name, ListPrice FROM Production.Product")
row = cursor.fetchone()
print(row.Name)   # Attribute access works directly
print(row[0])     # Index access also works

Error handling

The mssql-python driver uses the same exception hierarchy as pyodbc, so most exception handlers require only a module name change.

Exception hierarchy

The exception class names map directly between drivers:

pyodbc:

try:
    cursor.execute(query)
except pyodbc.Error as e:
    pass
except pyodbc.DatabaseError as e:
    pass
except pyodbc.OperationalError as e:
    pass

mssql-python:

try:
    cursor.execute("SELECT TOP 1 * FROM Production.Product")
    print(cursor.fetchone())
except mssql_python.Error as e:
    pass
except mssql_python.DatabaseError as e:
    pass
except mssql_python.OperationalError as e:
    pass

Error details

Both drivers expose error details through exception arguments:

pyodbc:

try:
    cursor.execute(query)
except pyodbc.Error as e:
    sqlstate = e.args[0]
    message = e.args[1]

mssql-python:

try:
    cursor.execute("SELECT TOP 1 * FROM NonExistentTable_XYZ")
except mssql_python.Error as e:
    # Error message contains SQLSTATE and details
    print(str(e))

Connection pooling

The mssql-python driver includes connection pooling by default, so external pooling libraries are no longer required.

Remove external pooling

If you used external pooling with pyodbc, the mssql-python driver has it built in:

Before (pyodbc external pool):

from dbutils.pooled_db import PooledDB

pool = PooledDB(pyodbc, 5, driver="{ODBC Driver 18 for SQL Server}",
                server="your_server", database="your_database",
                uid="your_username", pwd="your_password")
conn = pool.connection()

After (mssql-python built-in pooling):

conn = mssql_python.connect(connection_string)
conn.close()

Configure pool

Override the default pool size and timeout with mssql_python.pooling():

import mssql_python

mssql_python.pooling()

Complete migration example

The following shows the same function written with pyodbc and then rewritten with mssql-python.

Before (pyodbc)

This version uses the pyodbc connection string with a DRIVER keyword:

import pyodbc
from datetime import date

def get_orders(customer_id: int, start_date: date):
    conn = pyodbc.connect(
        "DRIVER={ODBC Driver 18 for SQL Server};"
        "SERVER=localhost;"
        "DATABASE=AdventureWorks2022;"
        "Trusted_Connection=yes;"
    )
    cursor = conn.cursor()

    cursor.execute("""
        SELECT SalesOrderID, OrderDate, TotalDue
        FROM Sales.SalesOrderHeader
        WHERE CustomerID = ? AND OrderDate >= ?
        ORDER BY OrderDate DESC
    """, (customer_id, start_date))

    orders = []
    for row in cursor:
        orders.append({
            "id": row.SalesOrderID,
            "date": row.OrderDate,
            "total": row.TotalDue
        })

    cursor.close()
    conn.close()
    return orders

After (mssql-python)

This version removes the DRIVER keyword. All queries, parameters, and row access patterns remain identical:

import mssql_python
from datetime import date

def get_orders(customer_id: int, start_date: date):
    conn = mssql_python.connect(
        "Server=localhost;"
        "Database=AdventureWorks2022;"
        "Trusted_Connection=yes;"
    )
    cursor = conn.cursor()

    cursor.execute("""
        SELECT SalesOrderID, OrderDate, TotalDue
        FROM Sales.SalesOrderHeader
        WHERE CustomerID = ? AND OrderDate >= ?
        ORDER BY OrderDate DESC
    """, (customer_id, start_date))

    orders = []
    for row in cursor:
        orders.append({
            "id": row.SalesOrderID,
            "date": row.OrderDate,
            "total": row.TotalDue
        })

    cursor.close()
    conn.close()
    return orders

The only changes are the import statement and the connection string (no DRIVER keyword needed). Every query, parameter, fetch pattern, and row access stays identical.

Testing migration

Before completing the migration, run the same queries against both drivers and compare results to confirm equivalent behavior.

Verify equivalent behavior

Use a comparison function that runs the same query against both drivers and asserts the results match:

import pyodbc
import mssql_python

def compare_results(pyodbc_conn_str: str, mssql_conn_str: str, query: str):
    """Compare results from both drivers."""
    # pyodbc query
    pyodbc_conn = pyodbc.connect(pyodbc_conn_str)
    pyodbc_cursor = pyodbc_conn.cursor()
    pyodbc_cursor.execute(query)
    pyodbc_results = pyodbc_cursor.fetchall()
    pyodbc_conn.close()
    
    # mssql-python query
    mssql_conn = mssql_python.connect(mssql_conn_str)
    mssql_cursor = mssql_conn.cursor()
    mssql_cursor.execute(query)
    mssql_results = mssql_cursor.fetchall()
    mssql_conn.close()
    
    # Compare
    assert len(pyodbc_results) == len(mssql_results)
    for p_row, m_row in zip(pyodbc_results, mssql_results):
        assert tuple(p_row) == tuple(m_row)
    
    print(f"Results match: {len(pyodbc_results)} rows")

Checklist

• [ ] Update imports from pyodbc to mssql_python.
• [ ] Remove DRIVER= from connection strings.
• [ ] Keep existing ? parameter queries (they work as-is).
• [ ] Use EXECUTE statements for stored procedure calls.
• [ ] Remove external connection pooling configuration.
• [ ] Update exception handling class names.
• [ ] Test all queries and stored procedures.
• [ ] Verify data type handling (especially decimals and dates).
• [ ] Remove ODBC driver from deployment requirements.

Related content

• Migrate from pymssql
• Installation
• Connection strings
• Executing queries
• Stored procedures