This MS SQL MCP Server provides an AI-friendly interface to explore and query your Microsoft SQL Server databases without coding. It exposes tools to discover database structure, inspect table details, perform safe read-only queries, and generate SQL from natural language prompts. The server supports a bundled interactive client and optional HTTP/SSE transport for network access, allowing Claude or other AI assistants to navigate tables, view schemas, and run constrained queries. Typical usage involves discovering tables, selecting a target table for details, running a read-only query, and iterating with natural language prompts to generate appropriate SQL queries. The system is designed to keep credentials local and queries read-only by default for safety.

Prerequisites:

• Node.js v14 or higher
• Access to a Microsoft SQL Server instance (on-premises or Azure)

Installation steps:

1. Clone the repository git clone https://github.com/dperussina/mssql-mcp-server.git
2. Navigate to the project directory cd mssql-mcp-server
3. Install dependencies npm install
4. Copy the example environment file cp .env.example .env
5. Configure your database connection in .env (example variables shown below) DB_USER=your_username DB_PASSWORD=your_password DB_SERVER=your_server_name_or_ip DB_DATABASE=your_database_name PORT=3333 HOST=0.0.0.0 TRANSPORT=stdio SERVER_URL=http://localhost:3333 DEBUG=false QUERY_RESULTS_PATH=/path/to/query_results
6. Start the server npm start

   or with HTTP/SSE transport

   npm run start:sse
7. (Optional) Run the interactive client npm run client

Notes and tips:

• This server emphasizes security by defaulting to read-only queries. To allow data modification, you would need to adjust query handling and permissions accordingly (not recommended for production AI access).
• Ensure your .env QUERY_RESULTS_PATH directory exists and is writable, as query results may be cached there.
• If you run into connection issues, verify DB_SERVER, DB_USER, and DB_PASSWORD are correct, and check DEBUG logs for details.
• The server can be exposed via HTTP/SSE transport for networked AI clients; adjust TRANSPORT and SERVER_URL accordingly.
• The included tools include:
  • mcp_SQL_mcp_discover_database(): Discover database structure
  • mcp_SQL_mcp_table_details({ tableName }): Inspect table schema
  • mcp_SQL_mcp_execute_query({ sql, returnResults }): Run a read-only SQL query and get results
  • mcp_SQL_mcp_discover_tables({ namePattern }): Find tables by name pattern
  • mcp_SQL_mcp_get_query_results({ uuid }): Retrieve saved large results
• For best results with AI prompting, combine discovery and table detail steps before crafting queries.