Microsoft Introduces SQL MCP Server: Secure Database Access for AI Agents
SQL MCP Server (1)

Microsoft has officially introduced the SQL MCP Server, a new tool designed to provide AI agents with secure, deterministic, and feature-rich access to enterprise data. A core component of Data API builder (DAB) 2.0, this server leverages the Model Context Protocol (MCP) to bridge the gap between AI models and production databases like SQL Server, Azure SQL, PostgreSQL, and MySQL.

Solving the "Fragile Query" Problem

Traditionally, connecting AI to databases relied on Natural Language to SQL (NL2SQL), which can be non-deterministic and prone to subtle, dangerous errors. SQL MCP Server takes a different approach called NL2DAB. It uses a well-defined entity abstraction layer and a built-in Query Builder to generate accurate T-SQL in a fully deterministic way, ensuring that the same natural language intent always produces the same safe result.

Core Features for the Enterprise

  • Zero-Code Deployment: Deployable as a simple container, SQL MCP Server allows developers to expose tables, views, and stored procedures using a single JSON configuration file.

  • Granular Security: It integrates with Role-Based Access Control (RBAC) and Azure Key Vault. Agents only see the data they are permitted to access, and internal schemas remain hidden via aliasing.

  • Performance & Scale: The server includes built-in L1 and L2 caching (with Redis support) and is fully instrumented with Azure Monitor, Application Insights, and OpenTelemetry.

  • DML-Only Safety: To protect production systems, the server focuses on Data Manipulation Language (DML)—reading, creating, and updating data—while intentionally omitting Data Definition Language (DDL) to prevent agents from altering the database schema.

The "Family of Seven" Tools

Instead of overwhelming an AI agent's context window with thousands of individual table definitions, SQL MCP Server exposes a fixed set of seven high-level tools:

  1. describe_entities: Discovery of available data objects.

  2. read_records: Querying and fetching data.

  3. create_record / update_record / delete_record: Standard CRUD operations.

  4. execute_entity: Running stored procedures (ideal for custom business logic).

  5. aggregate_records: Performing data calculations and summaries.

Developers can get started today by using the DAB CLI to initialize and run the server locally or in the cloud. For full technical documentation, visit aka.ms/sql/mcp.

Source: Microsoft