Oracle SQLcl version 25.4 is now available for download.
But practically, what’s it really mean for me?
I’m chatting with my Oracle AI Lakehouse (or data warehouse), and my queries take a LOOOONG time.
Some queries take awhile to run. Popular data warehouse benchmarking reports can take HOURS, even. But the default timeout for our MCP Server’s tool requests is only 60 seconds.
So what’s a business user to do?
Well now our run-sql tool allows you to request your LLM generated queries run asynchronously, or ‘in the background.’ This enabled an agent to ask for something in the database, and then come back later for the response. The agent is then free to tackle other parts of your plan, assuming they’re not dependent on the given query.
This also means it can task our MCP Server with multiple queries vs waiting/executing, one-by-one.
The MCP server response is immediate –
Task has been set to run in background successfully with id: 0
And this is all the information the Agent needs in order to request the results, later. Note there is no ‘new tool,’ for the Agent to interact with, it can simply invoke the run-sql tool again, this time asking for the status or results of job #0.
If this sounds familiar, then good on you! We added this feature in general for SQLcl in 2024 (24.1), and now the accompanying MCP Server follows the same workflow.
One more MCP thing…a ‘private’ set of connections
I’ll talk about this in more detail in a follow-up post, but you can tell SQLcl when it’s running as an MCP Server to use a specific directory for your settings and configuration. This effectively means you have a limited set of Oracle database connections available for AI Agents to use versus all of the connections you have already defined for SQLcl and SQL Developer Extension for VS Code.
Anything of interest in here for us humans?
Absolutely! The Help is easier to navigate.
Press space to go forward, B to go ‘back’, and use ‘/’ to enter a search term to jump to a topic/section.
Tab completion pretty much covers everything now.
Commands, sub-commands, help, filenames, table names, sql keywords, YES YES YES.
SQL History is more accessible, more search!
From the prompt, hit ‘Alt+R’ to activate search, enter your term, the first entry will pop into the buffer. use ‘Ctrl+R’ to execute it.
In addition, a new setting, AUTOSUGGESTIONS will ‘ghost’ the first matching query from your SQL History as you’re typing…
