mcp-mssql-server

A Model Context Protocol (MCP) server that provides a standardized interface for AI models to interact with MS SQL Server databases. This server implements the MCP specification to enable seamless database operations through a consistent API.

Features

- Execute SQL queries with parameter support
- List all tables in the database
- Describe table schemas
- Support for both stdio and HTTP transport modes
- Comprehensive logging system
- Environment-based configuration
- Error handling and graceful shutdown

Prerequisites

- Node.js (version that supports ES modules)
- MS SQL Server instance
- Access credentials for the database

Installation

1. Clone the repository
2. Install dependencies:
``bash
npm install
`

You can also install the package globally:
`bash
npm install -g mcp-mssql-server
`

3. Create a .env file in the project root with the following required variables:
`env
DB_SERVER=your_server_address
DB_USER=your_username
DB_PASSWORD=your_password
DB_DATABASE_NAME=your_database_name
`

Usage

The server can be started in two different transport modes:

$3


`bash
npm start

or

npm run start:stdio
`

$3

`bash
npm run start:http
`

$3

`bash
npm run dev
`

$3

The server implements the JSON-RPC 2.0 protocol with the following key methods:

1. initialize - Initialize the server connection:
`json
{
"jsonrpc": "2.0",
"id": 1,
"method": "initialize",
"params": {
"protocolVersion": "2024-11-05",
"capabilities": {
"tools": {}
},
"clientInfo": {
"name": "your-client",
"version": "1.0.0"
}
}
}
`

2. tools/list - List available tools
3. tools/call - Call a specific tool

Testing

The package includes a test client (test.package.js) that demonstrates how to interact with the MCP server:

`bash
node test.package.js
`

The test client implements an MCPTestClient class that:
- Spawns a server process using npx mcp-mssql-server
- Initializes the connection with protocol version '2024-11-05'
- Lists available tools
- Executes sample queries including:
- Listing all tables
- Running a specific SQL query to count documents
- Handles server responses and errors through stdio streams
- Includes proper error handling and process cleanup

Example test query from the client:
`json
{
"jsonrpc": "2.0",
"method": "tools/call",
"params": {
"name": "execute_sql_query",
"arguments": {
"query": "SELECT COUNT(*) as document_count FROM document_new"
}
}
}
`

The test client provides a simple way to verify the server's functionality and can serve as a reference for implementing your own client.

Available Tools

The server provides the following MCP tools:

$3

Execute SQL queries against the database with optional parameterization.

`json
{
"query": "SELECT * FROM Users WHERE id = @userId",
"parameters": [
{
"name": "userId",
"type": "int",
"value": 1
}
]
}
`

$3

List all available tables in the connected database.

$3

Get detailed schema information for a specific table.

`json
{
"table_name": "Users"
}
`

Logging

Logs are stored in the logs directory:
- error.log: Error-level logs
- combined.log`: All logs

In HTTP mode, logs are also output to the console.

Dependencies

Main dependencies:
- @modelcontextprotocol/sdk: ^1.13.0
- dotenv: ^16.4.5
- express: ^5.1.0
- mssql: ^11.0.1
- winston: ^3.11.0

Development dependencies:
- axios: ^1.10.0

License

MIT

Contributing

1. Fork the repository
2. Create your feature branch
3. Commit your changes
4. Push to the branch
5. Create a new Pull Request