Contents

1. Project
   1. Tech Stack
   2. Project structure
   3. LLM
   4. Running the application
   5. Sample Prompts
2. About my system, for reference
3. Issues
4. Inspirations / Documentation

Project

This project is a small demo for Spring AI.
It provides an MCP-Server to be used by your local LLM.
The tools provided by this application are:

• DateTimeTools:
  • Get the current time
  • Get the current date
  • Get the current year
  • Get the current week day

graph LR
;
    A(LLM) --> B(MCP-Server)
    B --> C(DateTimeTools)

Basically, this application is an aggregation layer for the LLM to get all the real-time data.

Tech Stack

• Java 21.0.8
• Maven 3.9.11
• Spring Boot 3.5.7
• Spring AI 1.1.0
• Docker Compose (Should work with Podman as well, just for the database)
• LM Studio 0.3.31

Project structure

Besides the standard Spring Boot Maven structure, we have some additional folder:

• docker: Contains the docker-compose.yml file to start the application and database locally

LLM

To use this project you need to run a LLM with tool support and probably thinking enabled; otherwise it did not work for me.
I recommend to use LM Studio with openai/gpt-oss-20b or qwen/qwen3-4b-thinking-2507 as the model.
openai/gpt-oss-20b has to be set to 16k context to load properly, but it is blazingly fast.
qwen/qwen3-4b-thinking-2507 is fast and has a small memory footprint.
To make sure you can see every setting, turn on Developer Mode.

MCP-Server configuration in LM Studio

On the right side under Program, Install and Edit mcp.json you can add this application here as a server:

{
  "mcpServers": {
    "showcase-lite": {
      "url": "http://localhost:58090/mcp"
    }
  }
}

That should be enough to get started. And the tools should be available as soon as you start the application.

LM Studio server configuration

To allow the application to connect to LM Studio for the chat in the application frontend, you need to navigate to Developer in the left menu.
Now you can toggle the server, and it should be available under http://localhost:1234. If you want to use LM Studio server in your whole network, you can toggle serve on local network in the server settings.
Please be aware that this server is not secure and should only be used in a local and trusted network.

Running the application

Sample Prompts

A valid answer for the following prompt should reflect your current local time, so no offsets or timezones.

What is the current time?

About my system, for reference

(no bragging, just for you to have an idea how the stuff might run on your system) I am using the following system:

• Linux Mint 22.2
• AMD Ryzen 7 9700X
• 64GB RAM, but not much is used for this project
• XFX 9070 XT 16GB

Performance:

Prompt	What is the current time? one tool call
qwen/qwen3-4b-thinking-2507	~2 seconds
ibm/granite-4-h-tiny	~1 second
openai/gpt-oss-20b (context reduced to 16k)	~1 second
microsoft/phi-4-reasoning-plus	did not terminate / endless loop in reasoning
mistralai/magistral-small-2509	~25 seconds
qwen/qwen3-14b	~50 seconds

Issues

• no differentiation between tools, resources, and prompts
• No tests / no useful tests
• performance is measured manually

Inspirations / Documentation

• Spring AI
• Github Badges