Search for a command to run...
"Bumpy roads lead to beautiful places"
No comments yet. Be the first to comment.
I saw Dimitri's post on SQL Assertions today and it piqued my curiosity. Putting more business rules closer to data is almost always a great idea, as you can now call DML from any technology and rest
Almost 20 years ago to the date, I submitted my resignation to Oracle to go start my own company. It was a difficult decision to make, as APEX - or HTML DB as it was called then - was just starting to gain traction. However, I was presented with an o...
One of the more difficult parts of the software development lifecycle is deployment. While it seems simple on the surface - just write some scripts and run them - it gets pretty complicated pretty quickly as the level of sophistication of your system...
When I first heard the term MCP used in the context of AI, I was immediately taken back to the 80s when the movie Tron was released. For those who have not seen it, Tron is a futuristic movie about a curious user who is literally sucked into a comput...
Another AI-centric tool that has raised a lot of interest is Claude Desktop. Like Cline, Claude Desktop uses the same reasoning engine at the core - Claude Code - to assist with development tasks. While they share the same engine, there are use cases for each.
Claude Desktop is a self-contained, simple interface. You interact with it outside of the scope of your IDE and code in a more general manner. While it can still generate some powerful solutions, Claude Desktop seems to be closer to an alternative to ChatGPT vs. a true AI coding assistant.
Cline is the better choice if you’re looking for an AI assistant that lives in your IDE. It’s there to help you write, review, and comment on code. It was designed for that single purpose from the ground up.
Another difference between Cline and Claude Desktop is how LLMs are integrated. Cline allows you to pick almost any LLM and use that. Thus, as Oracle, we have access to a handful, and can use any of them with Cline at no additional cost. Claude Desktop takes a different approach. You can use it for free for a limited number of questions per day, or you can sign up for one of the paid tiers.
Put another way - If I wanted to give an AI tool to my non-technical staff so that they could query an Oracle Database via MCP Server, it would be Claude Desktop without hesitation. Its simple, easy to understand interface and robust rendering capabilities make it an easy choice for the non-technical crowd.
Before we can use Claude Desktop, we need to configure it with the MCP Server and then connect to it. These steps take just minutes each.
Configuring Claude Desktop is as simple as configuring Cline. Simply follow the instructions in the SQLcl MCP Server documentation.
Once configured, you can validate that the MCP Server is running by navigating to Settings > Developer. You should see something like this:
Once configured, connecting to Oracle is done the same way that connecting to Cline is: just ask.
The SQL Developer connection apex_admin_user was created as a part of my last blog entry. Please have a look there if you missed it.
Now that we’re connected, let’s skip the easy stuff and jump right into asking Claude to build us some dashboards.
For our first dashboard, let’s use the data found in APEX_WORKSPACE_ACTIVITY_LOG - or page views. APEX will automatically track all page views - full and partial - in this view. Since our user has the APEX_ADMINISTRATOR_ROLE role, it will be able to see data across all of the workspaces. Keep in mind APEX will automatically purge data from the underlying table, so at best, you’ll have roughly two weeks of data.
Let’s start with the following prompt:
please create an analytics dashboard that summarizes page views for application 10000. include things like most popular page, avg render time, etc. make it look very slick and allow me to drill into details from the charts.
Like Cline, it required some back-and-forth to find the right view. Once it found it, it ran four queries to compute the data points that would ultimately end up in the dashboard. Remember - this dashboard is nothing more than static HTML with some Javascript.
Speaking of the dashboard it created, let’s have a look at it:
It’s pretty amazing that with just a couple of sentences, Claude was able to generate a professional looking dashboard that contains a variety of insights as to the page view data, including a combination of metrics, charts and a searchable report.
You can also drill into most of the data points. Since that’s harder to illustrate, feel free to try the dashboard for yourself here: https://claude.ai/public/artifacts/c022f815-7e81-4a16-b6f3-3cae15e82fd2
Next, let’s create a dashboard that reports on data that APEX doesn’t: user behavior. It would be interesting to analyze the same data set to see how users actually use the applications - specifically, which page flows are the most popular, which pages get used the most, etc.
Thus, let’s use this prompt to create it:
using the same page view data, come up with an assessment of common usage patterns based on session id. provide this data in a dashboard that provides insights as to what users do and what could be done to make the application better
This time, Claude didn’t need to think as much, as the source of the data was still in context from the previous dashboard. It got right to work and instantly created another dashboard that looks like this:
Wow! This one really impressed me, as not only did it provide a comprehensive dashboard with actionable data, but it also picked out some of the common user flows and trends in my applications and identified them.
Keep in mind that the dataset here is from an instance of APEX that I build and test things on. Thus, the usage patterns are more indicative of that versus actual usage across many users. Some of the recommendations reflect this.
Have a look at this one in action here: https://claude.ai/public/artifacts/5baab2fb-f0fe-4065-85d9-0d3844bae3bc
You may be thinking, “I thought this was an APEX blog, why are we creating Javascript applications all of a sudden?
Short answer - we can’t yet have AI generate APEX applications since there is not a published API to do so. Yet.
APEX Lang - not the existing APEX_LANG API, but the upcoming domain specific language with a similar name - will change that once it’s released. APEX Lang exposes all facets of an APEX application via APIs, allowing either users or AIs to call APIs to build and modify applications. Once it’s released, it will fundamentally change how we build and maintain APEX applications.
Claude Desktop enables anyone to generate content-rich, interactive dashboards without writing a single line of code. With minimal guidance, it even added insights and metrics to the dashboards, making them more than a bunch of reports & charts. Even if I simply harvested the SQL that Claude generated and used that in an APEX application, that’s a win.
I’m going to keep playing around with Claude Desktop and see what else I can come up with. Next, I’m going to dive into hooking up these dashboards to a RESTful endpoint hosted by ORDS so that they will always be accurate. Then I’ll look at how to secure the endpoint and perhaps allow for some updates to RESTful APIs.
Title photo by Mykhailo Amirdzhanian on Unsplash.
