mysql-mcp-pro
# mysql-mcp-pro
> 🇺🇸 [English version available here](README.en.md)
[](LICENSE)
[](https://www.python.org/downloads/)
[](https://modelcontextprotocol.io)
一个生产级的 **MySQL MCP Server**,让 AI 助手(Claude Desktop / WorkBuddy / Cursor / Cline 等)通过标准化协议安全高效地操作 MySQL 数据库。
## ✨ 为什么选 mysql-mcp-pro?
相比社区其他 MySQL MCP 实现,本项目提供了更完整的能力集:
| 特性 | mysql-mcp-pro | 其他 MCP |
|------|:---:|:---:|
| 连接池(减少握手开销) | ✅ | ❌ |
| 事务支持(begin/commit/rollback) | ✅ | ❌ |
| EXPLAIN 自动解读 | ✅ | ❌ |
| 结果导出 CSV/JSON(Excel 友好 BOM) | ✅ | ❌ |
| ER 关系图数据(nodes/edges) | ✅ | 少数 |
| 视图 / 存储过程 / 函数 / 触发器 浏览 | ✅ | 少数 |
| 查询历史 & 审计日志 | ✅ | ❌ |
| MCP Prompts 模板(生成/优化/对比/概览) | ✅ | ❌ |
| 安全分级(只读/DDL/危险语句/超时/行限/参数化) | ✅ | 基础 |
| 多传输模式(stdio / SSE / streamable-http) | ✅ | 部分 |
## 🚀 快速开始
### 1. 安装
```bash
# 从源码(推荐先 clone 再使用)
git clone https://github.com/CCCT173/mysql-mcp-pro.git
cd mysql-mcp-pro
pip install -e .
# 或从 PyPI(待发布)
pip install mysql-mcp-pro
```
### 2. 配置环境变量
复制 `.env.example` 为 `.env` 并编辑:
```env
MYSQL_HOST=localhost
MYSQL_PORT=3306
MYSQL_USER=root
MYSQL_PASSWORD=your_password
MYSQL_DATABASE=your_database
READ_ONLY=false # 只读模式
ALLOW_DDL=false # 是否允许 CREATE/ALTER/DROP
MAX_RESULT_ROWS=1000 # 单次查询最大返回行数
QUERY_TIMEOUT=30 # 查询超时(秒)
POOL_SIZE=5
AUDIT_LOG=true # 审计日志
EXPORT_DIR=./exports # 导出目录
```
### 3. 在你的 MCP 客户端注册
#### WorkBuddy
编辑 `~/.workbuddy/mcp.json`:
```json
{
"mcpServers": {
"mysql": {
"command": "python",
"args": ["-m", "server"],
"cwd": "D:/path/to/mysql-mcp-pro",
"env": {
"MYSQL_HOST": "localhost",
"MYSQL_PORT": "3306",
"MYSQL_USER": "root",
"MYSQL_PASSWORD": "your_password",
"MYSQL_DATABASE": "your_database"
}
}
}
}
```
或者如果你已经 `pip install -e .`:
```json
{
"mcpServers": {
"mysql": {
"command": "mysql-mcp-pro",
"env": {
"MYSQL_HOST": "localhost",
"MYSQL_PORT": "3306",
"MYSQL_USER": "root",
"MYSQL_PASSWORD": "your_password",
"MYSQL_DATABASE": "your_database"
}
}
}
}
```
#### Claude Desktop
编辑 `~/Library/Application Support/Claude/claude_desktop_config.json`(macOS)或 `%APPDATA%/Claude/claude_desktop_config.json`(Windows):
```json
{
"mcpServers": {
"mysql": {
"command": "mysql-mcp-pro",
"env": {
"MYSQL_HOST": "localhost",
"MYSQL_PORT": "3306",
"MYSQL_USER": "root",
"MYSQL_PASSWORD": "your_password",
"MYSQL_DATABASE": "your_database",
"READ_ONLY": "true"
}
}
}
}
```
#### Cursor / Cline
在项目根目录创建 `.cursor/mcp.json`(Cursor)或在 Cline 设置中添加,格式同上。
## Tools
> Tools are automatically extracted by MCP registries (Smithery, mcp.so, glama, etc.).
- **ping** — Test MySQL connection; returns server version and connection-pool stats.
- **get_server_status** — Key MySQL status metrics (connections, QPS, slow queries, InnoDB rows, traffic).
- **list_databases** — List all databases (system vs user).
- **list_tables** — List tables with engine, row count, size, comment. Optional `database` argument.
- **describe_table** — Show table schema: columns, primary key, indexes, foreign keys, CREATE SQL. Arguments: `table` (required), `database` (optional).
- **list_views** — List views in the database.
- **list_routines** — List stored procedures or functions. Arguments: `routine_type` (`PROCEDURE`/`FUNCTION`), `database`.
- **list_triggers** — List triggers in the database.
- **er_graph** — Cross-table foreign-key graph as nodes + edges, ready for ER diagrams.
- **execute_query** — Execute any SQL (read/write/DDL, safety-guarded), parameterized. Arguments: `sql`, `database`, `params`, `session_id`.
- **execute_select** — Read-only SELECT/SHOW/DESCRIBE/EXPLAIN with extra safety layer. Arguments: `sql`, `database`, `params`, `limit`.
- **get_table_data** — Quick paginated table fetch. Arguments: `table`, `columns`, `where`, `order_by`, `limit`, `offset`.
- **count_rows** — Count rows in a table with optional WHERE. Arguments: `table`, `database`, `where`.
- **explain_query** — Run `EXPLAIN` on a query and return plain-language performance insights. Arguments: `sql`, `database`, `params`, `format`.
- **export_query** — Export SELECT results to a CSV (UTF-8 BOM, Excel-friendly) or JSON file. Arguments: `sql`, `filename`, `format` (`csv`/`json`), `database`, `params`.
- **list_query_history** — Recent SQL executed in this session (duration, row count, errors). Argument: `limit`.
- **begin_transaction** — Start a new transaction; returns a `session_id` for chained calls. Argument: `database`.
- **commit_transaction** — Commit an open transaction. Argument: `session_id`.
- **rollback_transaction** — Rollback an open transaction. Argument: `session_id`.
- **show_processlist** — Show the current MySQL process list (`SHOW FULL PROCESSLIST`).
- **kill_query** — Kill a MySQL process (requires SUPER privilege). Argument: `process_id`.
## 🛠 工具一览(25 个)
### 连接 & 状态
- **`ping`** — 测试连接、返回服务器版本与连接池状态
- **`get_server_status`** — MySQL 关键指标(连接数、QPS、慢查询、流量等)
### 元数据浏览
- **`list_databases`** — 列出所有数据库(区分系统库 / 用户库)
- **`list_tables`** — 列出表,含引擎、行数、大小、注释
- **`describe_table`** — 表结构:字段、主键、索引、外键、建表 SQL
- **`list_views`** — 列出视图
- **`list_routines`** — 列出存储过程 / 函数
- **`list_triggers`** — 列出触发器
- **`er_graph`** — 跨表外键关系(nodes + edges),便于画 ER 图
### 查询执行
- **`execute_query`** — 执行任意 SQL(读/写/DDL,受安全策略约束),支持参数化
- **`execute_select`** — 只读 SELECT 专用(即使配置错误也只允许读)
- **`get_table_data`** — 快速取数:columns / where / order_by / limit / offset
- **`count_rows`** — 统计行数(支持可选 WHERE)
### 分析 & 导出
- **`explain_query`** — `EXPLAIN` 执行计划 + 中文性能洞察
- **`export_query`** — 结果导出 CSV(UTF-8 BOM,Excel 友好)或 JSON 文件
- **`list_query_history`** — 本会话 SQL 历史(耗时 / 行数 / 错误)
### 事务
- **`begin_transaction`** — 开启事务,返回 session_id
- **`commit_transaction`** — 提交事务
- **`rollback_transaction`** — 回滚事务
### 监控 & 管理
- **`show_processlist`** — 当前 MySQL 进程列表
- **`kill_query`** — Kill 指定进程(需要 SUPER 权限)
### MCP Resources(4 个)
- `mysql://databases` — 数据库列表
- `mysql://tables/{database}` — 表列表
- `mysql://schema/{database}/{table}` — 表结构
- `mysql://er/{database}` — ER 关系图数据
### MCP Prompts(4 个模板)
- **`generate_query`** — 根据自然语言和表信息生成 SQL
- **`optimize_sql`** — 分析并优化现有 SQL 性能
- **`compare_schema`** — 对比两张表的结构差异
- **`database_summary`** — 数据库整体概览报告
## 🔒 安全模型
```
危险语句(GRANT/REVOKE/SHUTDOWN/KILL/FLUSH/RESET)→ 永远禁止
DDL(CREATE/ALTER/DROP/RENAME/TRUNCATE) → 由 ALLOW_DDL 控制
写操作(INSERT/UPDATE/DELETE) → 由 READ_ONLY 控制
读操作(SELECT/SHOW/DESCRIBE/EXPLAIN) → 永远允许
```
- 参数化查询 `%s` 占位符,杜绝 SQL 注入
- 结果集默认 1000 行上限,防止意外返回海量数据
- 服务端 `MAX_EXECUTION_TIME` 强制查询超时
- 所有执行 SQL 落盘到 `exports/audit.log`
## 📁 项目结构
```
mysql-mcp-pro/
├── server.py # MCP 服务器(连接池 / 事务 / 导出 / 审计)
├── pyproject.toml # 包元数据,入口 mysql-mcp-pro
├── .env.example # 环境变量模板
├── LICENSE # MIT
├── README.md / README.en.md
└── scripts/
├── setup_test_db.py # 创建 mcp_test 样例数据库
├── test_v2.py # 53 项回归测试
└── diag.py # MySQL 握手包诊断工具
```
## 🧪 本地测试
```bash
# 1. 创建样例数据库(mcp_test,users + orders 表,带外键)
python scripts/setup_test_db.py
# 2. 跑全量回归测试
python scripts/test_v2.py
# → 53 passed, 0 failed
```
## 🤝 贡献
欢迎提 issue 和 PR!请确保:
1. 新功能带测试用例
2. 现有测试全部通过
3. 遵循现有代码风格
## 📄 License
[MIT](LICENSE)
TDQS
Scored across 21 tools
Each tool has a clear, distinct purpose. Ping and get_server_status differ; listing tools for databases, tables, views, routines, triggers are all separate. Query tools (execute_query, execute_select, get_table_data, count_rows, explain_query, export_query) each serve different needs. Process and transaction management tools are unambiguous.
All tool names use lowercase snake_case with a consistent verb_noun pattern (list_databases, describe_table, execute_query, begin_transaction, etc.). The only slight deviation is 'er_graph' which is a noun, but it still fits the descriptive style. Overall, naming is predictable and uniform.
21 tools cover a comprehensive range of database operations (connection, inspection, querying, export, process management, transactions). While on the higher side, each tool serves a specific purpose and the count is reasonable for a full-featured MySQL MCP server; slight reduction could be possible but not necessary.
The tool set covers most essential database operations: connection check, status, schema listing, query execution (with safety variants), analysis, export, process management, and transactions. Missing dedicated tools for schema DDL (create/alter/drop) and user management, but these can be accomplished via execute_query. Minor gaps but overall solid.