# Spring AI Model Context Protocol Demo Application

A demo application showcasing the integration of Spring AI with SQLite databases using the Model Context Protocol (MCP). This application enables natural language interactions with your SQLite database through a command-line interface.

It uses the [SQLite MCP-Server](https://github.com/modelcontextprotocol/servers/tree/main/src/sqlite) to enable running SQL queries, analyzing business data, and automatically generating business insight memos.

## Features

- Natural language querying of SQLite databases
- Predefined question mode for automated database analysis
- Seamless integration with OpenAI's language models
- Built on Spring AI and Model Context Protocol

## Prerequisites

- Java 17 or higher
- Maven 3.6+
- uvx package manager
- Git
- OpenAI API key
- SQLite (optional, for database modifications)

## Installation

1. Install uvx (Universal Package Manager):
   ```bash
   # Follow installation instructions at:
   https://docs.astral.sh/uv/getting-started/installation/
   ```

2. Clone the repository:
   ```bash
   git clone https://github.com/spring-projects/spring-ai-examples.git
   cd model-context-protocol/sqlite/simple
   ```

3. Set up your OpenAI API key:
   ```bash
   export OPENAI_API_KEY='your-api-key-here'
   ```

## Sample SQLite database

SQLite database files are portable across operating systems.  This repository contains a sample database file named `test.db`.

It has a `PRODUCTS` table and was created using the script `create-database.sh`

## Running the Application

### Predefined Questions
Runs through a set of preset database queries:
```bash
./mvnw spring-boot:run
```

## Architecture Overview

Spring AI's integration with MCP follows a simple chain of components:

1. **MCP Client** provides the base communication layer with your database
2. **Function Callbacks** expose database operations as AI-callable functions
3. **Chat Client** connects these functions to the AI model

The bean definitions are described below, starting with the `ChatClient`

### Chat Client

```java
@Bean
@Profile("!chat")
public CommandLineRunner predefinedQuestions(ChatClient.Builder chatClientBuilder,
                                           List<McpFunctionCallback> functionCallbacks,
                                           ConfigurableApplicationContext context) {
    return args -> {
        var chatClient = chatClientBuilder.defaultFunctions(functionCallbacks.toArray(new McpFunctionCallback[0]))
                .build();
         // Run Predefined Questions
         System.out.println(chatClient.prompt(
            "Can you connect to my SQLite database and tell me what products are available, and their prices?").call().content());
         // ...
    };
}
```

The chat client setup is remarkably simple - it just needs the function callbacks that were automatically created from the MCP tools. Spring's dependency injection handles all the wiring, making the integration seamless.

Now let's look at the other bean definitions in detail...

### Function Callbacks

The application registers MCP tools with Spring AI using function callbacks:

```java
@Bean
public List<McpFunctionCallback> functionCallbacks(McpSyncClient mcpClient) {
    return mcpClient.listTools(null)
            .tools()
            .stream()
            .map(tool -> new McpFunctionCallback(mcpClient, tool))
            .toList();
}
```

#### Purpose

This bean is responsible for:
1. Discovering available MCP tools from the client
2. Converting each tool into a Spring AI function callback
3. Making these callbacks available for use with the ChatClient


#### How It Works

1. `mcpClient.listTools(null)` queries the MCP server for all available tools
   - The `null` parameter represents a pagination cursor
   - When null, returns the first page of results
   - A cursor string can be provided to get results after that position
2. `.tools()` extracts the tool list from the response
3. Each tool is transformed into a `McpFunctionCallback` using `.map()`
4. These callbacks are collected into an array using `.toArray(McpFunctionCallback[]::new)`

#### Usage

The registered callbacks enable the ChatClient to:
- Access MCP tools during conversations
- Handle function calls requested by the AI model
- Execute tools against the MCP server (e.g., SQLite database)


### MCP Client 

The application uses a synchronous MCP client to communicate with the SQLite database:

```java
@Bean(destroyMethod = "close")
public McpSyncClient mcpClient() {
    var stdioParams = ServerParameters.builder("uvx")
            .args("mcp-server-sqlite", "--db-path", getDbPath())
            .build();

    var mcpClient = McpClient.sync(new StdioServerTransport(stdioParams),
            Duration.ofSeconds(10), new ObjectMapper());

    var init = mcpClient.initialize();
    System.out.println("MCP Initialized: " + init);

    return mcpClient;
}
```

This configuration:
1. Creates a stdio-based transport layer that communicates with the `uvx` MCP server
2. Specifies SQLite as the backend database and its location
3. Sets a 10-second timeout for operations
4. Uses Jackson for JSON serialization
5. Initializes the connection to the MCP server

The `destroyMethod = "close"` annotation ensures proper cleanup when the application shuts down.



## Documentation references

You can find out more about this sample application following the this quickstart link to a specific verison in github.

Unfortunately, on December 10th 2024, the quickstart was changed from SQLite to be a weather retrieval example.

However, here is the [link](https://github.com/modelcontextprotocol/docs/blob/1024e03f83aa0b8badde9b50dfee4d2e4e7f9446/quickstart.mdx) to the docs before the change if you want to read up on some details.

For example, you may want to create other tables and install SQLite