Two things separate this from the usual 'run SQL' server. The datasource list means the same connection does not have to be re-pointed when you move between environments — the model picks the database as part of the request. And the SQL safety policy is configurable rather than assumed, which is what makes handing an agent a write-capable connection defensible. If you need a database other than MySQL, the author points at a sibling project instead.
A Spring AI server that executes SQL against MySQL, holding several datasources at once and switching between them at runtime.
- Arbitrary SQL executed against a configured MySQL database
- Several datasources defined side by side in datasource.yml, one marked as the default
- Runtime switching between them, so one server covers development and staging without a restart
- SQL safety controls that stop a model from running destructive statements
- Groovy scripts as an extension point when you want behaviour beyond plain SQL
JDK 21 or higher and Maven 3.6 or higher. Connections are declared in datasource.yml as JDBC URLs with a username and password — there is no separate account to create. Launched either through the Maven wrapper or from a built jar; the -Dloader.path argument is only needed if you use the Groovy extensions. Documentation is primarily in Chinese, with an English README alongside it.