AutoMCP-SQL
Automatically scans SQLite databases and generates typed CRUD tools for each table, allowing natural language database interaction without exposing raw SQL.
README
AutoMCP-SQL — Zero-Knowledge MCP Server
An MCP server that autonomously scans any SQLite database, generates typed CRUD tools for every table it discovers, builds a schema-aware prompt explaining joins/relationships, and enforces Zero-Knowledge security — the LLM never writes raw SQL, only calls pre-validated parameterized templates.
What It Does
| Feature | Detail |
|---|---|
| Auto-Discovery | Scans sqlite_master at startup; no manual table registration |
| CRUD Generation | Creates get_, create_, update_, delete_ tools per table in memory |
| Schema Prompt | Injects table columns + foreign-key relationships as MCP instructions |
| Zero-Knowledge SQL | Instead of allowing the LLM to write arbitrary SQL (which risks SQL injection and destructive operations), the server generates isolated CRUD tools (Create, Read, Update, Delete) in memory for each table. The LLM interacts with these safe, pre-validated python templates |
| Configurable DB | Point at any SQLite file via DB_PATH env var |
Project Structure
AutoMCP-SQL/
├── server.py # MCP server — scans DB, generates tools, starts server
├── setupdb.py # One-time script to create and seed legacy.db for testing
├── legacy.db # SQLite database (auto-created by setupdb.py)
├── pyproject.toml # Dependencies managed by uv
├── uv.lock # Lockfile — guarantees reproducible installs
└── README.md
Tech Stack
| Tool | Why |
|---|---|
| FastMCP | Minimal decorator-based MCP server; zero boilerplate for tool registration |
| sqlite3 (stdlib) | No extra dependency; sufficient for parameterized CRUD on any SQLite file |
| uv | 10-100× faster than pip; lockfile guarantees reproducible installs across machines |
| MCP (Model Context Protocol) | Standard protocol for giving LLMs structured, auditable tool access |
Prerequisites
- Python 3.11+
- uv — fast Python package/project manager
- Claude Desktop — to connect the MCP server (optional)
Install uv if you don't have it:
# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
# Windows (PowerShell)
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
Setup
1. Clone the repository
git clone https://github.com/Mav977/AutoMCP-SQL.git
cd AutoMCP-SQL
2. Install dependencies
uv sync
This reads uv.lock and recreates the exact virtual environment.
3. Seed the demo database
uv run setupdb.py
This creates legacy.db in the project root with sample tables (users, orders, products, etc.) so you can test immediately.
Using your own database? Skip this step and point
DB_PATHat your existing.dbfile (see Configuration below).
Example: Claude Desktop Configuration
You can access the configuration file directly from within the app:
- Open Claude Desktop.
- Navigate to Settings.
- Select the Developer tab.
- Click Edit Config to open the
claude_desktop_config.jsonfile in your default text editor.
2. Add the MCP server entry
Replace the path below with the absolute path to your cloned server.py:
Windows:
{
"mcpServers": {
"AutoMCP-SQL": {
"command": "uv",
"args": [
"run",
"--with",
"mcp[cli]",
"mcp",
"run",
"C:\\Users\\YourName\\Desktop\\Projects\\AutoMCP-SQL\\server.py"
],
"env": {
"DB_PATH": "C:\\Users\\YourName\\Desktop\\Projects\\AutoMCP-SQL\\legacy.db"
}
}
}
}
macOS / Linux:
{
"mcpServers": {
"AutoMCP-SQL": {
"command": "uv",
"args": [
"run",
"--with",
"mcp[cli]",
"mcp",
"run",
"/home/yourname/projects/AutoMCP-SQL/server.py"
],
"env": {
"DB_PATH": "/home/yourname/projects/AutoMCP-SQL/legacy.db"
}
}
}
}
Important: Use double backslashes (
\\) in Windows paths inside JSON.
3. Restart Claude Desktop
Fully quit and reopen Claude Desktop. The MCP server starts automatically when Claude launches.
4. Verify the connection
Open a new conversation in Claude Desktop. Click the plus icon (+) in the chat input area, select Connectors, and then click Manage connectors. Here, you will see AutoMCP-SQL listed along with all the tools the MCP has access to (like get_users, create_orders, etc.).
You can also just ask Claude:
"What tables do you have access to?"
Claude will use the injected schema prompt to describe the database structure without querying it.
Using Your Own Database
Set DB_PATH in the config to point at any existing SQLite file:
"env": {
"DB_PATH": "C:\\path\\to\\your\\database.db"
}
The server will scan it on startup and auto-generate tools for every table it finds — no code changes needed.
Configuration
| Environment Variable | Default | Description |
|---|---|---|
DB_PATH |
./legacy.db (next to server.py) |
Path to the SQLite database file |
How Zero-Knowledge Security Works
The server never exposes raw SQL execution to the LLM. Instead:
- On startup,
scan_db()reads the schema and builds a natural-language prompt describing tables and relationships. - For each table, four tool functions are registered with hard-coded SQL templates:
get_<table>→SELECT * FROM <table> WHERE col = ?create_<table>→INSERT INTO <table> (cols) VALUES (?)update_<table>→UPDATE <table> SET col = ? WHERE id = ?delete_<table>→DELETE FROM <table> WHERE id = ?
- All values are passed as parameterized arguments — never string-interpolated into queries.
- Claude can only call these named tools; it has no mechanism to execute arbitrary SQL.
This means even a prompt-injected or jailbroken model cannot run DROP TABLE or exfiltrate data via UNION — the attack surface is limited to the four CRUD templates.
Troubleshooting
Server doesn't appear in Claude Desktop
- Wait a few seconds; it may take some time to load.
- Double-check the path in
claude_desktop_config.json— it must be the absolute path toserver.py, not a folder. - Make sure
uvis on your system PATH (open a new terminal and runuv --versionto verify). - Fully quit Claude Desktop (system tray on Windows, Cmd+Q on Mac) and reopen it.
DB_PATH does not exist error
Run uv run setupdb.py first to create legacy.db, or update DB_PATH in the config to point at an existing database.
Tools not showing up after config change Claude Desktop only reads the config on launch. Fully restart it after any config edits.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。