# 构建自定义 MCP 客户端

如果你在构建自己的 LLM 应用，可以直接使用官方 MCP SDK 连接到 Synapse MCP Server。这样你的应用就能访问全部 79 个 Synapse 工具。

## SDK

| 语言 | 包 |
|----------|---------|
| TypeScript/JavaScript | `@modelcontextprotocol/sdk` |
| Python | `mcp` |

## TypeScript 示例

### 安装

```bash
npm install @modelcontextprotocol/sdk
```

### 通过 stdio 连接

```typescript
import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";

const transport = new StdioClientTransport({
  command: "npx",
  args: ["-y", "synapse-mcp-api@latest"],
  env: {
    SYNAPSE_MIND_KEY: process.env.SYNAPSE_MIND_KEY!,
    SYNAPSE_URL: "https://synapse.schaefer.zone",
  },
});

const client = new Client(
  { name: "my-app", version: "1.0.0" },
  { capabilities: {} }
);

await client.connect(transport);

// 列出所有可用工具
const { tools } = await client.listTools();
console.log(`Available tools: ${tools.length}`);
for (const tool of tools) {
  console.log(`- ${tool.name}: ${tool.description}`);
}

// 调用工具
const result = await client.callTool({
  name: "memory_recall",
  arguments: {},
});
console.log(result.content);

// 存储记忆
await client.callTool({
  name: "memory_store",
  arguments: {
    category: "fact",
    key: "custom_client_test",
    content: "Built a custom MCP client",
    tags: ["test", "mcp"],
    priority: "normal",
  },
});

await client.close();
```

### 通过 HTTP/SSE 连接（远程）

```typescript
import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { SSEClientTransport } from "@modelcontextprotocol/sdk/client/sse.js";

const transport = new SSEClientTransport(
  new URL("https://synapse-mcp.schaefer.zone/sse"),
  {
    requestInit: {
      headers: {
        Authorization: `Bearer ${process.env.SYNAPSE_MIND_KEY}`,
      },
    },
  }
);

const client = new Client(
  { name: "my-app", version: "1.0.0" },
  { capabilities: {} }
);

await client.connect(transport);
// ... 同上使用
```

## Python 示例

### 安装

```bash
pip install mcp
```

### 通过 stdio 连接

```python
from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client

server_params = StdioServerParameters(
    command="npx",
    args=["-y", "synapse-mcp-api@latest"],
    env={
        "SYNAPSE_MIND_KEY": "mk_YOUR_KEY",
        "SYNAPSE_URL": "https://synapse.schaefer.zone",
    },
)

async with stdio_client(server_params) as (read, write):
    async with ClientSession(read, write) as session:
        await session.initialize()

        # 列出工具
        tools = await session.list_tools()
        print(f"Available tools: {len(tools.tools)}")

        # 调用工具
        result = await session.call_tool("memory_recall", {})
        print(result.content)

        # 存储记忆
        await session.call_tool("memory_store", {
            "category": "fact",
            "key": "python_client_test",
            "content": "Built a Python MCP client",
            "tags": ["test", "mcp", "python"],
            "priority": "normal",
        })
```

## 工具配置文件

连接时，可通过 `Mcp-Tool-Profile` 头（HTTP/SSE）或 `MCP_PROFILE` 环境变量（stdio）请求特定工具配置文件：

```typescript
// stdio：设置环境变量
env: {
  SYNAPSE_MIND_KEY: "mk_...",
  MCP_PROFILE: "minimal",  // 8 个工具而非 119 个
}

// HTTP/SSE：设置头
requestInit: {
  headers: {
    Authorization: "Bearer mk_...",
    "Mcp-Tool-Profile": "minimal",
  },
}
```

## 错误处理

```typescript
try {
  const result = await client.callTool({ name: "memory_recall", arguments: {} });
  if (result.isError) {
    console.error("Tool error:", result.content);
  } else {
    console.log("Success:", result.content);
  }
} catch (err) {
  console.error("MCP error:", err);
}
```

## 用例

- **自定义 AI 助手** — 构建带持久记忆的 Agent
- **工作流自动化** — 在自定义工作流中串联 Synapse 工具
- **数据流水线** — 提取记忆、转换、加载到别处
- **监控仪表板** — 展示记忆统计、聊天历史、任务

## 下一步

- [MCP 规范](https://spec.modelcontextprotocol.io)
- [Synapse MCP 仓库](https://gitlab.com/schaefer-services/synapse-mcp)
- [API 概览](/docs/api/overview)