---
description:
globs:
alwaysApply: false
---
# MCP SDK集成

本项目支持Model Context Protocol (MCP) SDK集成，允许通过标准化协议执行代码。MCP相关功能主要在[mcp.rs](mdc:src/mcp.rs)中实现。

## 核心数据结构

```rust
#[derive(Serialize, Deserialize, Debug)]
pub struct CodeExecutionParams {
    pub code: String,
    pub language: String,
    pub show_logs: bool,
}

#[derive(Serialize, Deserialize, Debug)]
pub struct CodeExecutionResult {
    pub logs: Vec<String>,
    pub result: Option<String>,
    pub error: Option<String>,
}
```

## JavaScript执行

MCP中的JavaScript执行使用以下包装代码：

```javascript
// MCP SDK integration for JavaScript
// Note: MCP SDK is not required for basic functionality

// Log capture setup
const logs = [];
const originalConsoleLog = console.log;
console.log = function() {
    const message = Array.from(arguments).map(arg => 
        typeof arg === 'object' && arg !== null ? JSON.stringify(arg) : String(arg)
    ).join(' ');
    logs.push(message);
    if (show_logs) {
        originalConsoleLog.apply(console, arguments);
    }
};

async function mcpExecute() {
    try {
        // User code here

        // Execute handler
        let result = null;
        if (typeof handler === 'function') {
            result = await handler();
        }

        return {
            logs: logs,
            result: result !== undefined ? String(result) : null,
            error: null
        };
    } catch (error) {
        return {
            logs: logs,
            result: null,
            error: error.toString()
        };
    }
}

// Execute and return results
mcpExecute().then(result => {
    originalConsoleLog(JSON.stringify(result));
});
```

## Python执行

MCP中的Python执行使用以下包装代码：

```python
# MCP SDK integration for Python
# Note: MCP SDK is not required for basic functionality
import sys
import json
import io
from contextlib import redirect_stdout

# Store logs
logs = []

# Create custom stdout to capture logs
class LogCapture(io.StringIO):
    def write(self, text):
        if text.strip():
            logs.append(text.rstrip())
            if show_logs:
                sys.__stdout__.write(text)

def mcp_execute():
    try:
        // User code here

        // Execute handler function and capture result
        result = None
        with redirect_stdout(LogCapture()):
            if 'handler' in globals() and callable(handler):
                result = handler()
        
        return {
            "logs": logs,
            "result": str(result) if result is not None else None,
            "error": None
        }
    except Exception as e:
        return {
            "logs": logs,
            "result": None,
            "error": str(e)
        }

# Execute and print results
result = mcp_execute()
print(json.dumps(result))
```

## 执行流程

MCP执行流程由[execute_with_mcp](mdc:src/mcp.rs#L138)函数处理：

1. 创建临时文件并写入包装后的代码
2. 根据语言类型选择执行命令（Deno或uv）
3. 执行代码并捕获输出
4. 解析JSON格式的执行结果
5. 如果无法解析JSON，则返回原始输出

## 使用方法

在命令行中使用`--use-mcp`参数启用MCP SDK集成：

```bash
cargo run -- --use-mcp js -c "function handler() { return 'Hello from MCP JS'; }"
cargo run -- --use-mcp python -f examples/test_python.py
```