从零到一：构建完整的MCP应用
本教程将带您从零开始，构建一个完整的MCP应用。我们将创建一个智能代码审查助手，它能够分析代码、检测问题、提供改进建议。
项目概述
功能特性
- ✅ 代码语法检查
- ✅ 代码风格分析
- ✅ 类型检查
- ✅ 安全漏洞检测
- ✅ 性能优化建议
- ✅ 与GitHub集成
技术栈
- 语言：Python 3.11+
- MCP SDK：mcp
- 代码分析：ast, pylint, mypy, bandit
- GitHub集成：PyGithub
- LLM：OpenAI API
第一步：环境准备
1.1 创建项目目录
mkdir mcp-code-reviewer
cd mcp-code-reviewer
1.2 创建虚拟环境
# 使用venv
python -m venv venv
# 激活虚拟环境
# Linux/Mac:
source venv/bin/activate
# Windows:
venv\Scripts\activate
1.3 安装依赖
创建 requirements.txt：
mcp>=0.1.0
pylint>=3.0.0
mypy>=1.5.0
bandit>=1.7.0
PyGithub>=1.59.0
openai>=1.0.0
python-dotenv>=1.0.0
安装依赖：
pip install -r requirements.txt
1.4 创建项目结构
mkdir -p src tests config
touch src/__init__.py
touch src/server.py
touch src/tools.py
touch src/utils.py
touch .env.example
touch README.md
第二步：配置环境变量
创建 .env 文件：
# GitHub配置
GITHUB_TOKEN=your_github_token_here
# OpenAI配置
OPENAI_API_KEY=your_openai_api_key_here
# MCP服务器配置
MCP_SERVER_NAME=code-reviewer
MCP_SERVER_VERSION=1.0.0
创建 .env.example：
GITHUB_TOKEN=
OPENAI_API_KEY=
MCP_SERVER_NAME=code-reviewer
MCP_SERVER_VERSION=1.0.0
第三步：实现核心功能
3.1 创建工具模块 (src/tools.py)
"""MCP工具实现模块"""
import ast
import subprocess
import tempfile
import os
from typing import List, Dict, Any
from github import Github
import openai
from dotenv import load_dotenv
load_dotenv()
g = Github(os.getenv("GITHUB_TOKEN"))
openai.api_key = os.getenv("OPENAI_API_KEY")
def analyze_code_syntax(code: str, language: str = "python") -> Dict[str, Any]:
"""分析代码语法"""
if language != "python":
return {"valid": True, "errors": []}
try:
ast.parse(code)
return {"valid": True, "errors": []}
except SyntaxError as e:
return {
"valid": False,
"errors": [{
"type": "syntax_error",
"message": e.msg,
"line": e.lineno,
"column": e.col_offset
}]
}
def analyze_code_style(code: str) -> Dict[str, Any]:
"""分析代码风格"""
with tempfile.NamedTemporaryFile(mode='w', suffix='.py', delete=False) as f:
f.write(code)
temp_file = f.name
try:
result = subprocess.run(
["pylint", temp_file, "--output-format=json"],
capture_output=True,
text=True,
timeout=30
)
import json
issues = json.loads(result.stdout) if result.stdout else []
return {
"score": calculate_style_score(issues),
"issues": issues[:10] # 只返回前10个问题
}
except Exception as e:
return {"score": 0, "issues": [{"message": str(e)}]}
finally:
os.unlink(temp_file)
def analyze_code_types(code: str) -> Dict[str, Any]:
"""类型检查"""
with tempfile.NamedTemporaryFile(mode='w', suffix='.py', delete=False) as f:
f.write(code)
temp_file = f.name
try:
result = subprocess.run(
["mypy", temp_file, "--show-error-codes"],
capture_output=True,
text=True,
timeout=30
)
return {
"valid": result.returncode == 0,
"errors": result.stdout.split("\n") if result.stdout else []
}
except Exception as e:
return {"valid": False, "errors": [str(e)]}
finally:
os.unlink(temp_file)
def analyze_security(code: str) -> Dict[str, Any]:
"""安全漏洞检测"""
with tempfile.NamedTemporaryFile(mode='w', suffix='.py', delete=False) as f:
f.write(code)
temp_file = f.name
try:
result = subprocess.run(
["bandit", "-f", "json", "-ll", temp_file],
capture_output=True,
text=True,
timeout=30
)
import json
report = json.loads(result.stdout) if result.stdout else {}
return {
"high_risk": len([i for i in report.get("results", []) if i.get("issue_severity") == "HIGH"]),
"medium_risk": len([i for i in report.get("results", []) if i.get("issue_severity") == "MEDIUM"]),
"issues": report.get("results", [])[:10]
}
except Exception as e:
return {"high_risk": 0, "medium_risk": 0, "issues": [{"message": str(e)}]}
finally:
os.unlink(temp_file)
def generate_suggestions(code: str, analysis_results: Dict[str, Any]) -> str:
"""使用LLM生成改进建议"""
syntax_valid = analysis_results.get('syntax', {}).get('valid', False)
style_score = analysis_results.get('style', {}).get('score', 0)
types_valid = analysis_results.get('types', {}).get('valid', False)
high_risk = analysis_results.get('security', {}).get('high_risk', 0)
medium_risk = analysis_results.get('security', {}).get('medium_risk', 0)
syntax_status = '通过' if syntax_valid else '失败'
types_status = '通过' if types_valid else '失败'
# 构建提示词，避免在f-string中使用嵌套代码块
code_block = f"```python\n{code}\n```"
prompt = f"""
分析以下Python代码，并提供改进建议：
代码：
{code_block}
分析结果：
- 语法检查：{syntax_status}
- 代码风格得分：{style_score}/10
- 类型检查：{types_status}
- 安全风险：高风险 {high_risk} 个，中风险 {medium_risk} 个
请提供具体的改进建议，包括：
1. 代码质量问题
2. 性能优化建议
3. 安全改进建议
4. 最佳实践建议
"""
try:
system_msg = "你是一个专业的代码审查助手。"
user_msg = prompt
# 构建消息列表
message_system = dict(role="system", content=system_msg)
message_user = dict(role="user", content=user_msg)
messages_list = [message_system, message_user]
response = openai.ChatCompletion.create(
model="gpt-4",
messages=messages_list,
temperature=0.7,
max_tokens=1000
)
return response.choices[0].message.content
except Exception as e:
return f"生成建议时出错：{str(e)}"
def calculate_style_score(issues: List[Dict]) -> float:
"""计算代码风格得分"""
if not issues:
return 10.0
# 根据问题数量和严重程度计算得分
error_count = len([i for i in issues if i.get("type") == "error"])
warning_count = len([i for i in issues if i.get("type") == "warning"])
score = 10.0 - (error_count * 0.5) - (warning_count * 0.1)
return max(0.0, min(10.0, score))
def review_github_pr(repo_name: str, pr_number: int) -> Dict[str, Any]:
"""审查GitHub Pull Request"""
try:
repo = g.get_repo(repo_name)
pr = repo.get_pull(pr_number)
review_comments = []
for file in pr.get_files():
if file.filename.endswith('.py'):
# 获取文件内容
content = repo.get_contents(file.filename, ref=pr.head.sha).decoded_content.decode()
# 分析代码
syntax_result = analyze_code_syntax(content)
style_result = analyze_code_style(content)
types_result = analyze_code_types(content)
security_result = analyze_security(content)
analysis = {
"file": file.filename, # 文件名
"syntax": syntax_result,
"style": style_result,
"types": types_result,
"security": security_result
}
review_comments.append(analysis)
return {
"pr_number": pr_number,
"repo": repo_name,
"files_reviewed": len(review_comments),
"reviews": review_comments
}
except Exception as e:
return {"error": str(e)}
3.2 创建MCP服务器 (src/server.py)
"""MCP服务器主文件"""
import asyncio
import os
from mcp.server import Server
from mcp.types import Tool, TextContent
from dotenv import load_dotenv
from .tools import (
analyze_code_syntax,
analyze_code_style,
analyze_code_types,
analyze_security,
generate_suggestions,
review_github_pr
)
load_dotenv()
app = Server(
name=os.getenv("MCP_SERVER_NAME", "code-reviewer"),
version=os.getenv("MCP_SERVER_VERSION", "1.0.0")
)
@app.list_tools()
async def list_tools() -> list[Tool]:
"""列出所有可用工具"""
return [
Tool(
name="analyze_code",
description="分析代码并提供详细报告，包括语法、风格、类型和安全检查",
inputSchema={
"type": "object",
"properties": {
"code": {
"type": "string",
"description": "要分析的代码"
},
"language": {
"type": "string",
"description": "编程语言，默认为python",
"default": "python"
}
},
"required": ["code"]
}
),
Tool(
name="review_github_pr",
description="审查GitHub Pull Request中的所有Python文件",
inputSchema={
"type": "object",
"properties": {
"repo": {
"type": "string",
"description": "GitHub仓库名称，格式：owner/repo"
},
"pr_number": {
"type": "integer",
"description": "Pull Request编号"
}
},
"required": ["repo", "pr_number"]
}
),
]
@app.call_tool()
async def call_tool(name: str, arguments: dict) -> list[TextContent]:
"""处理工具调用"""
try:
if name == "analyze_code":
code = arguments.get("code")
language = arguments.get("language", "python")
# 执行各种分析
syntax_result = analyze_code_syntax(code, language)
style_result = analyze_code_style(code)
types_result = analyze_code_types(code)
security_result = analyze_security(code)
# 生成综合报告
analysis_results = {
"syntax": syntax_result,
"style": style_result,
"types": types_result,
"security": security_result
}
# 生成改进建议
suggestions = generate_suggestions(code, analysis_results)
# 格式化报告
syntax_valid = syntax_result.get('valid', False)
style_score = style_result.get('score', 0)
style_issues_count = len(style_result.get('issues', []))
types_valid = types_result.get('valid', False)
types_errors = types_result.get('errors', [])[:5]
high_risk = security_result.get('high_risk', 0)
medium_risk = security_result.get('medium_risk', 0)
syntax_status = '✅ 通过' if syntax_valid else '❌ 失败'
types_status = '✅ 通过' if types_valid else '❌ 失败'
types_errors_str = chr(10).join(types_errors)
report = f"""
# 代码分析报告
## 语法检查
{syntax_status}
{syntax_result.get('errors', [])}
## 代码风格
得分：{style_score}/10
问题数量：{style_issues_count}
## 类型检查
{types_status}
{types_errors_str}
## 安全分析
高风险：{high_risk} 个
中风险：{medium_risk} 个
## 改进建议
{suggestions}
"""
return [TextContent(type="text", text=report)]
elif name == "review_github_pr":
repo = arguments.get("repo")
pr_number = arguments.get("pr_number")
result = review_github_pr(repo, pr_number)
if "error" in result:
error_msg = result['error']
return [TextContent(type="text", text=f"错误：{error_msg}")]
# 格式化PR审查报告
repo_name = result['repo']
pr_num = result['pr_number']
files_count = result['files_reviewed']
report = f"""
# GitHub PR审查报告
**仓库**：{repo_name}
**PR编号**：#{pr_num}
**审查文件数**：{files_count}
## 文件审查详情
"""
for review in result.get('reviews', []):
review_file = review['file']
review_syntax_valid = review['syntax'].get('valid', False)
review_style_score = review['style'].get('score', 0)
review_types_valid = review['types'].get('valid', False)
review_high_risk = review['security'].get('high_risk', 0)
syntax_icon = '✅' if review_syntax_valid else '❌'
types_icon = '✅' if review_types_valid else '❌'
report += f"""
### {review_file}
- 语法：{syntax_icon}
- 风格得分：{review_style_score}/10
- 类型：{types_icon}
- 安全风险：高风险 {review_high_risk} 个
"""
return [TextContent(type="text", text=report)]
else:
unknown_tool_msg = f"未知工具：{name}"
return [TextContent(type="text", text=unknown_tool_msg)]
except Exception as e:
error_msg = f"执行工具时出错：{str(e)}"
return [TextContent(type="text", text=error_msg)]
async def main():
"""启动MCP服务器"""
from mcp.server.stdio import stdio_server
async with stdio_server() as (read_stream, write_stream):
await app.run(
read_stream,
write_stream,
app.create_initialization_options()
)
if __name__ == "__main__":
asyncio.run(main())
第四步：配置MCP客户端
4.1 Claude Desktop配置
编辑 ~/Library/Application Support/Claude/claude_desktop_config.json (Mac) 或 %APPDATA%\Claude\claude_desktop_config.json (Windows)：
{
"mcpServers": {
"code-reviewer": {
"command": "python",
"args": [
"/path/to/mcp-code-reviewer/src/server.py"
],
"env": {
"GITHUB_TOKEN": "your_token",
"OPENAI_API_KEY": "your_key"
}
}
}
}
第五步：测试应用
5.1 本地测试
创建测试文件 tests/test_tools.py：
import pytest
from src.tools import analyze_code_syntax, analyze_code_style
def test_syntax_analysis():
"""测试语法分析"""
valid_code = "def hello():\n print('Hello')"
result = analyze_code_syntax(valid_code)
assert result["valid"] == True
invalid_code = "def hello(\n print('Hello')"
result = analyze_code_syntax(invalid_code)
assert result["valid"] == False
def test_style_analysis():
"""测试风格分析"""
code = "def hello():\n print('Hello')"
result = analyze_code_style(code)
assert "score" in result
assert 0 <= result["score"] <= 10
运行测试：
pytest tests/
5.2 使用测试
在Claude Desktop中测试：
用户：分析这段代码：
def calculate(x, y):
return x+y
Claude：调用analyze_code工具分析代码...
第六步：部署
6.1 创建Dockerfile
FROM python:3.11-slim
WORKDIR /app
# 安装系统依赖
RUN apt-get update && apt-get install -y \
gcc \
&& rm -rf /var/lib/apt/lists/*
# 安装Python依赖
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
# 复制代码
COPY src/ ./src/
COPY .env.example .env
# 运行服务器
CMD ["python", "-m", "src.server"]
6.2 创建docker-compose.yml
version: '3.8'
services:
mcp-code-reviewer:
build: .
environment:
- GITHUB_TOKEN=${GITHUB_TOKEN}
- OPENAI_API_KEY=${OPENAI_API_KEY}
volumes:
- ./data:/app/data
restart: unless-stopped
第七步：优化和扩展
7.1 添加缓存
from functools import lru_cache
import hashlib
@lru_cache(maxsize=100)
def cached_analyze_code(code_hash: str):
"""缓存代码分析结果"""
# 分析逻辑...
pass
7.2 添加日志
import logging
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger(__name__)
@app.call_tool()
async def call_tool(name: str, arguments: dict):
logger.info(f"调用工具：{name}")
# ...
总结
通过本教程，您已经：
- ✅ 创建了完整的MCP应用
- ✅ 实现了代码分析功能
- ✅ 集成了GitHub API
- ✅ 添加了LLM改进建议
- ✅ 配置了测试和部署