mcp-db-connect 1.0.4

mcp-db-connect

.NET global tool to expose SQL databases as a professional MCP server.

Installation

dotnet tool install --global mcp-db-connect

Quick Start (as global tool)

mcp-db-connect --sqlserver --host <HOST> --database <DB> --user <USER> --password <PASS> [options]

You can use the same arguments for MySQL, PostgreSQL, or SQLite.

---

Usage examples with different database engines

SQL Server

mcp-db-connect --sqlserver --host localhost --database testdb --user sa --password password --port 1433 --trustServerCertificate true

MySQL

mcp-db-connect --mysql --host localhost --database testdb --user user --password password --port 3306

PostgreSQL

mcp-db-connect --postgresql --host localhost --database testdb --user user --password password --port 5432

SQLite

mcp-db-connect --sqlite --database C:/path/to/your/file.db

---

Local development execution

You can build and run the server directly from the source code, useful for development and debugging:

dotnet build
dotnet run --project MCP_db -- --sqlserver --host <HOST> --database <DB> --user <USER> --password <PASS> [options]

Examples:

# SQL Server
dotnet run --project MCP_db -- --sqlserver --host localhost --database testdb --user sa --password password --port 1433 --trustServerCertificate true

# MySQL
dotnet run --project MCP_db -- --mysql --host localhost --database testdb --user user --password password --port 3306

# PostgreSQL
dotnet run --project MCP_db -- --postgresql --host localhost --database testdb --user user --password password --port 5432

# SQLite
dotnet run --project MCP_db -- --sqlite --database C:/path/to/your/file.db

---

Integration with editors and MCP clients

Configure your editor (VS Code, Cursor, etc.) to use the global or local command. Here are examples for each engine:

SQL Server

{
  "sqlserverCSharpMCP": {
    "command": "mcp-db-connect",
    "args": [
      "--sqlserver",
      "--host", "SERVER",
      "--instance", "INSTANCE",
      "--database", "db_name",
      "--port", "1433",
      "--user", "user",
      "--password", "password",
      "--trustServerCertificate", "true"
    ]
  }
}

MySQL

{
  "mysqlCSharpMCP": {
    "command": "mcp-db-connect",
    "args": [
      "--mysql",
      "--host", "mysql_host",
      "--database", "db_name",
      "--port", "3306",
      "--user", "user",
      "--password", "password"
    ]
  }
}

PostgreSQL

{
  "postgresCSharpMCP": {
    "command": "mcp-db-connect",
    "args": [
      "--postgresql",
      "--host", "postgres_host",
      "--database", "db_name",
      "--port", "5432",
      "--user", "user",
      "--password", "password"
    ]
  }
}

SQLite

{
  "sqliteCSharpMCP": {
    "command": "mcp-db-connect",
    "args": [
      "--sqlite",
      "--database", "C:/path/to/your/file.db"
    ]
  }
}

Or, for local development:

{
  "sqlserverCSharpMCP": {
    "command": "dotnet",
    "args": [
      "run", "--project", "MCP_db", "--", "--sqlserver", "--host", "SERVER", "--database", "db_name", "--user", "user", "--password", "password"
    ]
  }
}

⚠️ Do not include real credentials in shared configuration files or repositories. Use environment variables or local files ignored by git for sensitive data.

---

Functional example: interaction with an AI agent and MCP tools

Suppose you have an AI agent (e.g., in VS Code, Cursor, or an LLM) connected to the MCP server:

User:

Show me the tables in the database

AI Agent:

• Detects the intent and automatically calls the MCP tool list_tables.
• The server executes the query and responds with the list of tables.

Agent's response:

The tables in the database are: clientes, ordenes, productos.

---

Another example:

User:

How many records are in the clientes table?

AI Agent:

• Calls the MCP tool read_query with the query SELECT COUNT(*) FROM clientes.
• The server executes the query and responds with the result.

Agent's response:

The clientes table has 1245 records.

---

Export example:

User:

Export the data from the productos table in CSV format

AI Agent:

• Calls the MCP tool export_query with the query SELECT * FROM productos and format csv.
• The server generates the CSV and delivers it to the user.

Agent's response:

Here is the CSV file with the data from the productos table.

---

Example: search and show a stored procedure

User:

Find the stored procedure usp_ObtenerClientes and show me its content

AI Agent:

• Calls the MCP tool find_stored_procedure with the name usp_ObtenerClientes.
• The server searches for the procedure and retrieves its SQL definition.

Agent's response:

The stored procedure usp_ObtenerClientes has the following content:

CREATE PROCEDURE usp_ObtenerClientes
AS
SELECT * FROM clientes

---

These flows work the same for SQL Server, MySQL, PostgreSQL, or SQLite, and can be used from any compatible MCP client.

---

What is MCP and what is it for?

• MCP is an open protocol that allows applications and editors to interact with tools and structured data in a standard, secure, and extensible way.
• This MCP server exposes database operations (queries, administration, insights) as MCP tools, which can be consumed by compatible clients (VS Code, Cursor, LLMs, etc.).

---

Troubleshooting and best practices

• Do not write logs to stdout: Only the MCP protocol (JSON) should go to stdout. Use the .NET logger or Console.Error for logs.
• Extensibility: To add new tools, create static methods in a class decorated with [McpServerToolType] and use [McpServerTool].
• Layer separation: Keep business logic in Application, technical implementations in Infrastructure, and configuration in the main project.
• Validation: Always validate input parameters and the connection string.
• If the MCP client shows JSON errors, make sure there are no logs in stdout.
• If tools do not appear, check that they are correctly decorated and that the assembly is registered with .WithToolsFromAssembly.
• Use stderr logs for advanced debugging.

---

Credits and License

• Based on the official Model Context Protocol SDK.
• Architecture inspired by Clean Architecture.
• MIT License.