Skip to content

Commit

Permalink
add mcp examples
Browse files Browse the repository at this point in the history
  • Loading branch information
chickenlj committed Dec 30, 2024
1 parent ff958ac commit 204b733
Show file tree
Hide file tree
Showing 34 changed files with 2,644 additions and 0 deletions.
1 change: 1 addition & 0 deletions pom.xml
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,7 @@
<module>spring-ai-alibaba-prompt-example</module>
<module>spring-ai-alibaba-rag-example</module>
<module>spring-ai-alibaba-structured-example</module>
<module>spring-ai-alibaba-mcp-example</module>
</modules>

<dependencyManagement>
Expand Down
3 changes: 3 additions & 0 deletions spring-ai-alibaba-mcp-example/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
# Spring AI Alibaba Example

The filesystem and sqlite example are originally from the [spring-ai](https://github.com/spring-projects/spring-ai) project.
Binary file not shown.
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
distributionUrl=https://repo.maven.apache.org/maven2/org/apache/maven/apache-maven/3.8.4/apache-maven-3.8.4-bin.zip
wrapperUrl=https://repo.maven.apache.org/maven2/org/apache/maven/wrapper/maven-wrapper/3.1.0/maven-wrapper-3.1.0.jar
119 changes: 119 additions & 0 deletions spring-ai-alibaba-mcp-example/filesystem/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,119 @@
# Spring AI 智能体通过 MCP 集成本地文件数据

这里我们提供一个示例智能体应用,这个智能体可以通过 MCP 查询或更新本地文件系统,并以文件系统中的数据作为上下文与模型交互。次示例演示如何使用模型上下文协议(MCP)将 Spring AI 与本地文件系统进行集成。

## 运行示例
### 前提条件
1. 安装 npx (Node Package eXecute):
首先确保本地机器安装了 [npm](https://docs.npmjs.com/downloading-and-installing-node-js-and-npm),然后运行如下命令:

```bash
npm install -g npx
```

2. 下载示例源码

```bash
git clone https://github.com/springaialibaba/spring-ai-alibaba-examples.git
cd spring-ai-alibaba-examples/spring-ai-alibaba-mcp-example/filesystem
```

3. 设置环境变量

```bash
# 通义大模型 Dashscope API-KEY
export AI_DASHSCOPE_API_KEY=${your-api-key-here}
# 要集成的本地数据所在文件目录
export MCP_FILE_DIRECTORY_PATH=${path-to-example}/spring-ai-alibaba-examples/spring-ai-alibaba-mcp-example/filesystem/data
```

4. 构建示例

```bash
./mvnw clean install
```



### 运行示例应用
运行示例,智能体将向模型发起提问(源码中包含预置问题,可通过源码查看),可通过控制台查看输出结果。

```bash
./mvnw spring-boot:run
```

## 示例架构(源码说明)
前文中我们讲解了 Spring AI 与 MCP 集成的基础架构,在接下来的示例中,我们将用到以下关键组件:

1. **MCP Client**,与 MCP 集成的关键,提供了与本地文件系统进行交互的能力。
2. **Function Callbacks**,Spring AI MCP 的 function calling 声明方式。
3. **Chat Client**,Spring AI 关键组件,用于LLM模型交互、智能体代理。

### 声明 ChatClient
```java
// List<McpFunctionCallback> functionCallbacks;
var chatClient = chatClientBuilder.defaultFunctions(functionCallbacks).build();
```

和开发之前的 Spring AI 应用一样,我们先定义一个 ChatClient Bean,用于与大模型交互的代理。需要注意的是,我们为 ChatClient 注入的 functions 是通过 MCP 组件(McpFunctionCallback)创建的。

接下来让我们具体看一下 McpFunctionCallback 是怎么使用的。

### 声明 MCP Function Callbacks
以下代码段通过 `mcpClient`与 MCP server 交互,将 MCP 工具通过 McpFunctionCallback 适配为标准的 Spring AI function。

1. 发现 MCP server 中可用的工具 tool(Spring AI 中叫做 function) 列表
2. 依次将每个 tool 转换成 Spring AI function callback
3. 最终我们会将这些 McpFunctionCallback 注册到 ChatClient 使用



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



可以看出,ChatClient 与模型交互的过程是没有变化的,模型在需要的时候告知 ChatClient 去做函数调用,只不过 Spring AI 通过 McpFunctionCallback 将实际的函数调用过程委托给了 MCP,通过标准的 MCP 协议与本地文件系统交互:

+ 在与大模交互的过程中,ChatClient 处理相关的 function calls 请求
+ ChatClient 调用 MCP 工具(通过 McpClient)
+ McpClient 与 MCP server(即 filesystem)交互

### 初始化 McpClient
该智能体应用使用同步 MCP 客户端与本地运行的文件系统 MCP server 通信:



```java
@Bean(destroyMethod = "close")
public McpSyncClient mcpClient() {
var stdioParams = ServerParameters.builder("npx")
.args("-y", "@modelcontextprotocol/server-filesystem", "path))
.build(); // 1
var mcpClient = McpClient.sync(new StdioServerTransport(stdioParams),
Duration.ofSeconds(10), new ObjectMapper()); //2
var init = mcpClient.initialize(); // 3
System.out.println("MCP Initialized: " + init);
return mcpClient;
}
```
在以上代码中:
1. 配置 MCP server 启动命令与参数
2. 初始化 McpClient:关联 MCP server、指定超时时间等
3. Spring AI 会使用 `npx -y @modelcontextprotocol/server-filesystem "/path/to/file"`在本地机器创建一个独立的子进程(代表本地 Mcp server),Spring AI 与 McpClient 通信,McpClient 进而通过与 Mcp server 的连接操作本地文件。
Original file line number Diff line number Diff line change
@@ -0,0 +1,20 @@
Model Context Protocol (MCP) Java SDK

A Java implementation of the Model Context Protocol specification, providing both synchronous and
asynchronous clients for MCP server interactions.

Overview: This SDK implements the Model Context Protocol, enabling seamless integration with AI models and tools through
a standardized interface. It supports both synchronous and asynchronous communication patterns, making it suitable for
various use cases and integration scenarios.

Features

Synchronous and Asynchronous client implementations
Standard MCP operations support:
Tool discovery and execution
Resource management and templates
Prompt handling and management
Resource subscription system
Server initialization and ping
Stdio-based server transport
Reactive programming support using Project Reactor
10 changes: 10 additions & 0 deletions spring-ai-alibaba-mcp-example/filesystem/data/summary.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,10 @@
# Summary of Spring AI MCP Overview

The `spring-ai-mcp-overview.txt` file provides an overview of the Model Context Protocol (MCP) Java SDK, which is a Java implementation of the MCP specification. It offers both synchronous and asynchronous clients for interacting with MCP servers.

## Key Features

- **Client Implementations:** The SDK supports both synchronous and asynchronous client implementations.
- **Standard MCP Operations:** It includes support for standard MCP operations such as tool discovery and execution, resource management, prompt handling, and server initialization.
- **Transport Mechanism:** The SDK uses a stdio-based server transport mechanism.
- **Reactive Programming:** It has reactive programming support using Project Reactor.
Loading

0 comments on commit 204b733

Please sign in to comment.