MSSQL MCP Server
# MSSQL MCP Server
A Model Context Protocol (MCP) server that enables LLMs like Claude to interact with Microsoft SQL Server databases through natural language.
## Features
- 🔍 Query your SQL Server database using natural language
- 📊 Read, insert, update, and delete data
- 🏗️ Create and manage tables and indexes
- 🔒 Secure connection handling with optional read-only mode
- ⚡ Direct TypeScript execution with tsx - no build step required
## Quick Start
### Option 1: Use directly from GitHub with npx (Recommended)
No installation needed! Just configure Claude Desktop:
#### Windows
Add to `%APPDATA%\Claude\claude_desktop_config.json`:
```json
{
"mcpServers": {
"mssql": {
"type": "stdio",
"command": "cmd",
"args": ["/c", "npx", "-y", "github:EvilPhatBoi/McpSqlServer"],
"env": {
"SERVER_NAME": "your-server.database.windows.net",
"DATABASE_NAME": "your-database",
"SQL_USERNAME": "your-username",
"SQL_PASSWORD": "your-password",
"PORT": "1433",
"ENCRYPT": "true",
"TRUST_SERVER_CERTIFICATE": "false",
"CONNECTION_TIMEOUT": "30",
"READONLY": "false"
}
}
}
}
```
#### macOS/Linux
Add to `~/Library/Application Support/Claude/claude_desktop_config.json`:
```json
{
"mcpServers": {
"mssql": {
"type": "stdio",
"command": "npx",
"args": ["-y", "github:EvilPhatBoi/McpSqlServer"],
"env": {
"SERVER_NAME": "your-server.database.windows.net",
"DATABASE_NAME": "your-database",
"SQL_USERNAME": "your-username",
"SQL_PASSWORD": "your-password",
"PORT": "1433",
"ENCRYPT": "true",
"TRUST_SERVER_CERTIFICATE": "false",
"CONNECTION_TIMEOUT": "30",
"READONLY": "false",
"DEBUG": "false"
}
}
}
}
```
**Note for legacy SQL Servers:** If you experience SSL/TLS connection issues with older SQL Server versions, try:
- `"ENCRYPT": "false"` to disable TLS/SSL encryption
- `"TRUST_SERVER_CERTIFICATE": "true"` to trust self-signed certificates
- `"CONNECTION_TIMEOUT": "60"` or higher for slower connections
- `"DEBUG": "true"` to see detailed connection logs
### Option 2: Clone and run locally
1. Clone the repository:
```bash
git clone https://github.com/EvilPhatBoi/McpSqlServer.git
cd McpSqlServer
```
2. Install dependencies:
```bash
npm install
```
3. Create `.env` file:
```bash
cp .env.example .env
# Edit .env with your database credentials
```
4. Configure Claude Desktop to point to your local installation:
```json
{
"mcpServers": {
"mssql": {
"type": "stdio",
"command": "npx",
"args": ["tsx", "C:/path/to/McpSqlServer/src/index.ts"],
"env": {
"SERVER_NAME": "your-server.database.windows.net",
"DATABASE_NAME": "your-database",
"SQL_USERNAME": "your-username",
"SQL_PASSWORD": "your-password"
}
}
}
}
```
## Environment Variables
| Variable | Description | Default |
|----------|-------------|---------|
| `SERVER_NAME` | SQL Server hostname | Required |
| `DATABASE_NAME` | Database name | Required |
| `SQL_USERNAME` | SQL username | Required |
| `SQL_PASSWORD` | SQL password | Required |
| `PORT` | SQL Server port | `1433` |
| `ENCRYPT` | Enable TLS/SSL encryption | `true` |
| `TRUST_SERVER_CERTIFICATE` | Trust self-signed certificates | `false` |
| `CONNECTION_TIMEOUT` | Connection timeout in seconds | `30` |
| `READONLY` | Enable read-only mode | `false` |
| `DEBUG` | Enable debug logging | `false` |
## Usage Examples
Once configured, you can interact with your database using natural language in Claude:
- "Show me all customers from New York"
- "Create a table called products with columns for id, name, and price"
- "Update the price of product with id 5 to 29.99"
- "List all tables in the database"
- "Describe the structure of the orders table"
## Development
### Running locally with tsx:
```bash
npm run start # Run the server
npm run dev # Run with watch mode
```
### Type checking:
```bash
npm run typecheck
```
## Security Notes
- Never commit `.env` files with real credentials
- Use read-only mode (`READONLY=true`) in production for safety
- The server requires WHERE clauses for updates to prevent accidental mass updates
- Consider using environment-specific credentials
## Troubleshooting
### Connection issues
- Ensure your SQL Server allows remote connections
- Check firewall rules for SQL Server port (default 1433, or your custom PORT setting)
- Verify credentials and server name
- Enable debug logging by setting `DEBUG=true` in your environment variables
### SSL/TLS connection issues
- For older SQL Server versions that don't support modern TLS, set `ENCRYPT=false`
- If using self-signed certificates, set `TRUST_SERVER_CERTIFICATE=true`
- For connection timeout issues, increase `CONNECTION_TIMEOUT` to 60 or higher
- Check that your SQL Server accepts the TLS version your Node.js supports
### Debug mode
To enable detailed logging for troubleshooting:
```json
"env": {
"DEBUG": "true",
// ... other environment variables
}
```
### Authentication errors
- This server uses SQL authentication, not Windows authentication
- Ensure SQL authentication is enabled on your server
- Check that the SQL user has appropriate permissions
## License
MIT
TDQS
Scored across 8 tools
Each tool has a clearly distinct purpose with no ambiguity. create_index, create_table, describe_table, drop_table, insert_data, list_table, read_data, and update_data each target specific database operations with no overlap in functionality. The descriptions clearly differentiate between schema operations, data manipulation, and querying.
All tools follow a consistent verb_noun pattern with snake_case throughout. The naming convention is perfectly uniform: create_index, create_table, describe_table, drop_table, insert_data, list_table, read_data, and update_data all maintain the same structure and style.
With 8 tools, this server is well-scoped for MSSQL database operations. Each tool earns its place by covering essential database tasks: schema management (create/drop/describe/list tables, create index) and data operations (insert/read/update). The count is appropriate for the domain without being overwhelming.
The tool surface covers most essential MSSQL operations well, including table lifecycle management and core CRUD operations. However, there are minor gaps: no delete_data tool (only update_data with WHERE clause), no transaction management tools, and no stored procedure or view operations. Agents can work around these gaps using existing tools.