mcp-mssql

# MCP SQL Server Tool

<!-- mcp-name: io.github.alyiox/mcp-mssql -->

[![Build Status](https://github.com/alyiox/mcp-mssql/actions/workflows/ci.yml/badge.svg?branch=main)](https://github.com/alyiox/mcp-mssql/actions/workflows/ci.yml)
[![NuGet Version](https://img.shields.io/nuget/v/Alyio.McpMssql.svg)](https://www.nuget.org/packages/Alyio.McpMssql)

A read-only-by-default [Model Context Protocol (MCP)](https://modelcontextprotocol.io) server for Microsoft SQL Server that supports metadata discovery, parameterized queries, and query analysis, with profile-based configuration. The query tools enforce SELECT-only (no DML/DDL); an optional `run_command` tool can execute arbitrary write T-SQL, but only on profiles that explicitly opt in via `AllowWrite` (locked off by default).

**Requirements:** .NET 8.0 or later runtime (the tool targets `net8.0` and `net10.0`), SQL Server, and a connection string. Building from source requires the .NET 10.0 SDK.

## Quick start

Set `MCPMSSQL_CONNECTION_STRING` and run the server in one of these ways:

```bash
# Option 1: Run from NuGet package (e.g. with MCP Inspector)
export MCPMSSQL_CONNECTION_STRING="Server=127.0.0.1;User ID=sa;Password=<YourStrong@Passw0rd>;Encrypt=True;TrustServerCertificate=True;"
npx -y @modelcontextprotocol/inspector dotnet dnx Alyio.McpMssql --prerelease
```

```bash
# Option 2: Install and run as a global tool
dotnet tool install --global Alyio.McpMssql --prerelease
export MCPMSSQL_CONNECTION_STRING="Server=127.0.0.1;User ID=sa;Password=<YourStrong@Passw0rd>;Encrypt=True;TrustServerCertificate=True;"
npx -y @modelcontextprotocol/inspector mcp-mssql
```

```bash
# Option 3: Run from source (clone repo, then)
export MCPMSSQL_CONNECTION_STRING="Server=127.0.0.1;User ID=sa;Password=<YourStrong@Passw0rd>;Encrypt=True;TrustServerCertificate=True;"
npx -y @modelcontextprotocol/inspector dotnet run --project src/Alyio.McpMssql
```

Use `--prerelease` for pre-release builds.

## Configuration

All settings use the **MCPMSSQL** prefix. **Flat** environment variables (e.g. `MCPMSSQL_CONNECTION_STRING`) are the straightforward way to configure the **default** profile when you have a single connection. For multiple profiles, the user-scoped `appsettings.json` file is recommended.

**Single connection:** Configure via environment variables.

```bash
# Connection string (required).
export MCPMSSQL_CONNECTION_STRING="Server=127.0.0.1;User ID=sa;Password=<YourStrong@Passw0rd>;Encrypt=True;TrustServerCertificate=True;"

# Optional description for the default profile (tooling/AI discovery).
export MCPMSSQL_DESCRIPTION="Primary connection"

# Optional max rows per interactive query (default `500`; hard ceiling `1000`).
export MCPMSSQL_QUERY_MAX_ROWS="500"

# Optional query timeout in seconds (default `30`).
export MCPMSSQL_QUERY_COMMAND_TIMEOUT_SECONDS="60"

# Optional max rows for snapshot queries (default `10000`; hard ceiling `50000`).
export MCPMSSQL_QUERY_SNAPSHOT_MAX_ROWS="10000"

# Optional snapshot query timeout in seconds (default `120`).
export MCPMSSQL_QUERY_SNAPSHOT_COMMAND_TIMEOUT_SECONDS="120"

# Optional analyze timeout in seconds (default `300`).
export MCPMSSQL_ANALYZE_COMMAND_TIMEOUT_SECONDS="300"

# Optional: enable write commands (DDL/DML) via run_command (default `false`).
# Soft guard only — prefer a db_datareader login for a hard read-only guarantee.
export MCPMSSQL_ALLOW_WRITE="false"

# Optional write command timeout in seconds (default `60`; hard ceiling `600`).
export MCPMSSQL_WRITE_COMMAND_TIMEOUT_SECONDS="60"
```

**Multiple connections:** Use the user-scoped `appsettings.json` file (recommended). Env vars also work via .NET host conventions (`MCPMSSQL__PROFILES__<NAME>__CONNECTIONSTRING`, etc.).

- Unix-like: `~/.config/mcp-mssql/appsettings.json`
- Windows: `%USERPROFILE%\.config\mcp-mssql\appsettings.json`

Example (`appsettings.json`):

```json
{
  "McpMssql": {
    "Profiles": {
      "default": {
        "ConnectionString": "Server=...;User ID=...;Password=...;",
        "Description": "Primary connection",
        "Query": {
          "MaxRows": 500,
          "CommandTimeoutSeconds": 60,
          "SnapshotMaxRows": 10000,
          "SnapshotCommandTimeoutSeconds": 120
        },
        "Analyze": {
          "CommandTimeoutSeconds": 300
        }
      },
      "warehouse": {
        "ConnectionString": "Server=warehouse.example.com;...",
        "Description": "Warehouse read-only"
      },
      "migrations": {
        "ConnectionString": "Server=...;User ID=...;Password=...;",
        "Description": "Write-enabled profile for schema changes",
        "AllowWrite": true,
        "Write": {
          "CommandTimeoutSeconds": 60
        }
      }
    }
  }
}
```

**Local development:** Store the connection string in user-secrets, then run with `DOTNET_ENVIRONMENT=Development` so secrets load.

```bash
dotnet user-secrets set "MCPMSSQL_CONNECTION_STRING" "..." --project src/Alyio.McpMssql
npx -y @modelcontextprotocol/inspector -e DOTNET_ENVIRONMENT=Development dotnet run --project src/Alyio.McpMssql
```

**Azure SQL / Microsoft Entra ID:** This MCP server uses [Microsoft.Data.SqlClient](https://www.nuget.org/packages/Microsoft.Data.SqlClient), which supports Microsoft Entra (Azure AD) authentication. Set the `Authentication` property in the connection string to a supported mode (e.g. `Active Directory Default`, `Active Directory Managed Identity`, or `Active Directory Interactive`) when connecting to Azure SQL. See [Connect to Azure SQL with Microsoft Entra authentication and SqlClient](https://learn.microsoft.com/en-us/sql/connect/ado-net/sql/azure-active-directory-authentication) for all modes and details.

## Tools and resources

All tools accept an optional `profile`; when omitted, the default profile is used.

**Tools**

| Tool | Description | Key params |
|---|---|---|
| **`list_profiles`** | List configured connection profiles. Call first when picking a non-default profile. | — |
| **`get_server_properties`** | Get server properties and execution limits (timeouts, row caps, guardrails). | `profile` |
| **`list_objects`** | List catalog metadata. `kind=catalog`: databases; `schema`: schemas; `relation`: tables/views; `routine`: procedures/functions. `catalog` omitted → active catalog (ignored for `kind=catalog`). `schema` omission depends on kind. | `kind`, `profile`, `catalog`, `schema` |
| **`get_object`** | Get metadata for one relation or routine. Use `list_objects` to resolve names. Returns empty detail payloads if `includes` is null. | `kind`, `name`, `profile`, `catalog`, `schema`, `includes` |
| **`run_query`** | Execute read-only T-SQL SELECT; only SELECT allowed (no DML/DDL). Returns results as CSV in the `data` field (inline) or a snapshot resource URI when `snapshot=true`. Inline limit: 500 rows (hard ceiling 1000). Snapshot limit: 10 000 rows. Prefer `analyze_query` for plan tuning. | `sql`, `profile`, `catalog`, `parameters`, `snapshot` |
| **`analyze_query`** | Analyze execution plan for a read-only SELECT. Returns compact JSON summary (cost, operators, cardinality, warnings, indexes, waits, stats). Fetch full XML from `plan_uri`; does not return result rows. | `sql`, `profile`, `catalog`, `parameters`, `estimated` |
| **`run_command`** | Execute write T-SQL (DDL/DML). Rejected unless the target `profile` sets `AllowWrite=true` (off by default). Caller manages transactions. Returns `rows_affected` (−1 for DDL) and server `messages`. Marked destructive; intended for human-supervised use. | `sql`, `profile`, `catalog`, `parameters` |

- **`kind`** — `catalog`, `schema`, `relation`, or `routine`. For `get_object`, only `relation` or `routine`.
- **`includes`** — Array of detail sections: `columns`, `indexes`, `constraints` (relations only), `definition` (routines only).

**Resources**

| URI template | Description |
|---|---|
| `mssql://profiles` | List configured connection profiles. Same data as `list_profiles`. |
| `mssql://server-properties?{profile}` | Get server properties and execution limits. Same data as `get_server_properties`. |
| `mssql://objects?{kind,profile,catalog,schema}` | List catalog metadata. Schema omission behavior matches `list_objects`. |
| `mssql://objects/{kind}/{name}{?profile,catalog,schema,includes}` | Get metadata for one relation or routine. `includes` is required. |
| `mssql://plans/{id}` | Retrieve full XML execution plan by ID from `analyze_query`; entries expire after 7 days. |
| `mssql://snapshots/{id}` | Retrieve full query result as CSV by ID from `run_query` (snapshot=true); entries expire after 1 day. |

Resources mirror their corresponding tools and return JSON (except `mssql://plans/{id}` which returns XML and `mssql://snapshots/{id}` which returns CSV).

## Security

The query tools (`run_query`, `analyze_query`) are read-only (`SELECT` only) and use parameterized `@paramName` binding. Use environment variables or user-secrets for connection strings—never commit secrets.

**Writes are opt-in.** The `run_command` tool executes arbitrary T-SQL. It is rejected unless the target profile sets `AllowWrite=true`, which defaults to `false`, so existing deployments stay read-only with no change. The tool is always advertised and rejects at call time on locked profiles.

`AllowWrite` is a soft, application-level guard, **not** a security boundary — it constrains this server, not the database. For a genuine read-only guarantee, connect with a login restricted to `db_datareader`, and keep write-enabled profiles pointed at credentials scoped to only what they need. `run_command` is marked `destructive` via MCP tool annotations so hosts can gate it behind confirmation, but honor those annotations at the host's discretion.

## MCP host examples

Snippets for common MCP clients. Replace the connection string with your own; ensure `dotnet` is on your PATH. The `env` block is not required if the connection string is already set via `appsettings.json` or environment variables.

### Cursor