# mcp-client

**Repository Path**: snow-sun/mcp-client

## Basic Information

- **Project Name**: mcp-client
- **Description**: 测试使用
- **Primary Language**: Java
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-11-20
- **Last Updated**: 2025-12-03

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# MCP Client

This project serves as a client for integrating with Large Language Models (LLMs) and connecting to the MCP Spring Repeater service.

## Features

1. Integration with Alibaba Cloud Bailian (DashScope) Qwen models
2. Connection to the MCP Spring Repeater service
3. RESTful APIs for interacting with both services
4. SSE (Server-Sent Events) streaming support for real-time responses
5. Tool calling capability to allow LLMs to invoke MCP services
6. Dynamic tool loading from local tool service

## Project Structure

```
mcp-client/
├── src/
│   ├── main/
│   │   ├── java/
│   │   │   └── com/mcp/client/
│   │   │       ├── config/          # Configuration classes
│   │   │       ├── controller/      # REST controllers
│   │   │       ├── service/         # Business logic services
│   │   │       └── McpClientApplication.java  # Main application class
│   │   └── resources/
│   │       ├── application.yml      # Application configuration
│   │       └── static/              # Static resources (HTML test page)
│   └── test/                       # Unit and integration tests
├── pom.xml                         # Maven configuration
└── README.md                       # Project documentation
```

## Configuration

The application can be configured via `application.yml`. Key configuration options include:

- `app.llm.openai-api-key`: Your DashScope API key (default: sk-6ea152938bea4279a82eb5908baabebe)
- `app.llm.openai-base-url`: Base URL for DashScope API (default: https://dashscope.aliyuncs.com/compatible-mode/v1)
- `app.llm.model`: Model to use (default: qwen-plus)
- `app.repeater.base-url`: Base URL for the MCP Spring Repeater service
- `app.repeater.api-key`: API key for the MCP Spring Repeater service
- `app.local-tools.url`: URL for the local tool service (default: http://127.0.0.1:7777/mcp-short/tools/list)

These values can also be overridden with environment variables.

## Building and Running

To build the project:

```bash
mvn clean package
```

To run the project:

```bash
mvn spring-boot:run
```

Or after building:

```bash
java -jar target/mcp-client-1.0-SNAPSHOT.jar
```

## APIs

### Chat Completion (Regular)

POST `/api/mcp/chat`

Send a prompt to the LLM and receive a complete response.

Example:
```bash
curl -X POST http://localhost:8080/api/mcp/chat \
  -H "Content-Type: application/json" \
  -d "Hello, how are you?"
```

### Chat Completion (With Tool Calling)

POST `/api/mcp/chat-with-tools`

Send a prompt to the LLM and receive a response, with the ability for the LLM to call MCP services when needed.

Example:
```bash
curl -X POST http://localhost:8080/api/mcp/chat-with-tools \
  -H "Content-Type: application/json" \
  -d "Please use the MCP service to process this request: {\"action\": \"get_user_info\", \"user_id\": 123}"
```

### Chat Completion (Streaming with SSE)

POST `/api/mcp/sse/chat`

Send a prompt to the LLM and receive a streaming response using Server-Sent Events (SSE).

Example:
```bash
curl -X POST http://localhost:8080/api/mcp/sse/chat \
  -H "Content-Type: application/json" \
  -d "Write a short poem about technology"
```

### Repeater Service

POST `/api/mcp/repeater`

Forward data to the MCP Spring Repeater service.

Example:
```bash
curl -X POST http://localhost:8080/api/mcp/repeater \
  -H "Content-Type: application/json" \
  -d '{"message": "Hello from MCP Client"}'
```

## Tool Calling

The tool calling feature allows the LLM to automatically invoke your MCP services when needed. The system provides:

1. A predefined tool called `call_mcp_service` which forwards requests to your MCP service through the repeater
2. Dynamic tools loaded from your local tool service at `http://127.0.0.1:7777/mcp-short/tools/list`

When you use the `/api/mcp/chat-with-tools` endpoint, the LLM can decide to call your MCP service by generating a tool call with a payload. The system will automatically execute the tool call and provide the result back to the LLM to generate a final response.

## Testing SSE Functionality

After starting the application, visit `http://localhost:8080` in your browser to access a test page with buttons for both regular and streaming requests.

## Dependencies

- Spring Boot 3.3.0
- OkHttp HTTP client
- Java 17+