Local MySQL Read-only MCP Server

This project provides a small stdio MCP server for reading local MySQL data. It is designed to run with:

D:\anaconda\envs\torch\python.exe

The server implements a minimal MCP JSON-RPC stdio layer directly because the official Python MCP SDK requires a newer Python than the current torch environment.

Install Dependency

Recommended:

conda install -n torch -c conda-forge pymysql

Fallback:

D:\anaconda\envs\torch\python.exe -m pip install PyMySQL -i https://pypi.tuna.tsinghua.edu.cn/simple

If both commands fail because of local SSL or proxy settings, install PyMySQL manually into the torch conda environment and then run the tests below.

Configure MySQL

Copy the example file and fill in local credentials:

Copy-Item mysql_mcp_config.example.json mysql_mcp_config.json

mysql_mcp_config.json is ignored by git. You can also point to another config file with:

$env:MYSQL_MCP_CONFIG = "D:\path\to\mysql_mcp_config.json"

Use a MySQL user with read-only permissions whenever possible. The MCP server only exposes read tools, but database permissions should still enforce the same rule.

MCP Client Configuration

Example stdio configuration:

{
  "mcpServers": {
    "local-mysql-readonly": {
      "command": "D:\\anaconda\\envs\\torch\\python.exe",
      "args": ["D:\\桌面\\aienglish\\server.py"],
      "env": {
        "MYSQL_MCP_CONFIG": "D:\\桌面\\aienglish\\mysql_mcp_config.json"
      }
    }
  }
}

Tools

• list_databases: list visible databases, excluding system schemas by default.
• list_tables: list tables and views in one database.
• describe_table: return column and index metadata.
• read_rows: read rows using structured filters, ordering, limit, and offset.

read_rows does not accept raw SQL. Database, table, and column names are validated against MySQL metadata, and values are sent as query parameters.

Allowed filter operators:

=, !=, <, <=, >, >=, LIKE, IN, IS NULL, IS NOT NULL

Test

D:\anaconda\envs\torch\python.exe -m unittest

After mysql_mcp_config.json is configured and PyMySQL is installed, validate manually with your MCP client by calling:

1. list_databases
2. list_tables
3. describe_table
4. read_rows with limit set to 5