392 lines
13 KiB
Python
392 lines
13 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
Local LLM MCP Server for Claude Code Integration
|
|
Claude Code → MCP Server → Local LLM (Qwen2.5-Coder)
|
|
"""
|
|
|
|
import asyncio
|
|
import json
|
|
import logging
|
|
import requests
|
|
import subprocess
|
|
import os
|
|
from pathlib import Path
|
|
from typing import Dict, List, Any, Optional
|
|
from mcp.server import Server
|
|
from mcp.types import (
|
|
Tool,
|
|
TextContent,
|
|
Resource,
|
|
PromptMessage,
|
|
GetPromptResult
|
|
)
|
|
|
|
# ログ設定
|
|
logging.basicConfig(level=logging.INFO)
|
|
logger = logging.getLogger("local-llm-mcp")
|
|
|
|
class LocalLLMServer:
|
|
def __init__(self, model: str = "qwen2.5-coder:14b-instruct-q4_K_M"):
|
|
self.model = model
|
|
self.ollama_url = "http://localhost:11434"
|
|
self.conversation_history = []
|
|
|
|
def call_ollama(self, prompt: str, system_prompt: str = "") -> str:
|
|
"""Ollamaにリクエストを送信"""
|
|
try:
|
|
full_prompt = f"{system_prompt}\n\nUser: {prompt}\nAssistant:"
|
|
|
|
response = requests.post(
|
|
f"{self.ollama_url}/api/generate",
|
|
json={
|
|
"model": self.model,
|
|
"prompt": full_prompt,
|
|
"stream": False,
|
|
"options": {
|
|
"temperature": 0.1,
|
|
"top_p": 0.95,
|
|
"num_predict": 2048,
|
|
"stop": ["User:", "Human:"]
|
|
}
|
|
},
|
|
timeout=60
|
|
)
|
|
|
|
if response.status_code == 200:
|
|
return response.json()["response"].strip()
|
|
else:
|
|
return f"Error: {response.status_code} - {response.text}"
|
|
|
|
except Exception as e:
|
|
logger.error(f"Ollama call failed: {e}")
|
|
return f"Connection error: {e}"
|
|
|
|
def get_project_context(self) -> str:
|
|
"""現在のプロジェクトの情報を取得"""
|
|
context = []
|
|
|
|
# 現在のディレクトリ
|
|
cwd = os.getcwd()
|
|
context.append(f"Current directory: {cwd}")
|
|
|
|
# Git情報
|
|
try:
|
|
git_status = subprocess.run(
|
|
["git", "status", "--porcelain"],
|
|
capture_output=True, text=True, cwd=cwd
|
|
)
|
|
if git_status.returncode == 0:
|
|
context.append(f"Git status: {git_status.stdout.strip() or 'Clean'}")
|
|
except:
|
|
context.append("Git: Not a git repository")
|
|
|
|
# ファイル構造(簡略版)
|
|
try:
|
|
files = []
|
|
for item in Path(cwd).iterdir():
|
|
if not item.name.startswith('.') and item.name not in ['node_modules', '__pycache__']:
|
|
if item.is_file():
|
|
files.append(f"📄 {item.name}")
|
|
elif item.is_dir():
|
|
files.append(f"📁 {item.name}/")
|
|
|
|
if files:
|
|
context.append("Project files:")
|
|
context.extend(files[:10]) # 最初の10個まで
|
|
|
|
except Exception as e:
|
|
context.append(f"File listing error: {e}")
|
|
|
|
return "\n".join(context)
|
|
|
|
# MCPサーバーのセットアップ
|
|
app = Server("local-llm-mcp")
|
|
llm = LocalLLMServer()
|
|
|
|
@app.tool("code_with_local_llm")
|
|
async def code_with_local_llm(
|
|
task: str,
|
|
include_context: bool = True,
|
|
model_override: str = ""
|
|
) -> str:
|
|
"""
|
|
ローカルLLMでコーディングタスクを実行
|
|
|
|
Args:
|
|
task: 実行したいコーディングタスク
|
|
include_context: プロジェクトコンテキストを含めるか
|
|
model_override: 使用するモデルを一時的に変更
|
|
"""
|
|
logger.info(f"Executing coding task: {task}")
|
|
|
|
# モデルの一時変更
|
|
original_model = llm.model
|
|
if model_override:
|
|
llm.model = model_override
|
|
|
|
try:
|
|
# システムプロンプト構築
|
|
system_prompt = """You are an expert coding assistant. You can:
|
|
1. Write, analyze, and debug code
|
|
2. Explain programming concepts
|
|
3. Suggest optimizations and best practices
|
|
4. Generate complete, working solutions
|
|
|
|
Always provide:
|
|
- Clear, commented code
|
|
- Explanations of your approach
|
|
- Any assumptions you've made
|
|
- Suggestions for improvements
|
|
|
|
Format your response clearly with code blocks and explanations."""
|
|
|
|
# プロジェクトコンテキストを追加
|
|
if include_context:
|
|
context = llm.get_project_context()
|
|
system_prompt += f"\n\nCurrent project context:\n{context}"
|
|
|
|
# LLMに送信
|
|
response = llm.call_ollama(task, system_prompt)
|
|
|
|
return response
|
|
|
|
except Exception as e:
|
|
logger.error(f"Code generation failed: {e}")
|
|
return f"❌ Error in code generation: {e}"
|
|
finally:
|
|
# モデルを元に戻す
|
|
llm.model = original_model
|
|
|
|
@app.tool("read_file_with_analysis")
|
|
async def read_file_with_analysis(
|
|
filepath: str,
|
|
analysis_type: str = "general"
|
|
) -> str:
|
|
"""
|
|
ファイルを読み込んでLLMで分析
|
|
|
|
Args:
|
|
filepath: 分析するファイルのパス
|
|
analysis_type: 分析タイプ (general, bugs, optimization, documentation)
|
|
"""
|
|
logger.info(f"Analyzing file: {filepath}")
|
|
|
|
try:
|
|
# ファイル読み込み
|
|
with open(filepath, 'r', encoding='utf-8', errors='ignore') as f:
|
|
content = f.read()
|
|
|
|
# 分析タイプに応じたプロンプト
|
|
analysis_prompts = {
|
|
"general": "Analyze this code and provide a general overview, including its purpose, structure, and key components.",
|
|
"bugs": "Review this code for potential bugs, errors, or issues. Suggest fixes if found.",
|
|
"optimization": "Analyze this code for performance optimizations and suggest improvements.",
|
|
"documentation": "Generate comprehensive documentation for this code, including docstrings and comments."
|
|
}
|
|
|
|
prompt = f"{analysis_prompts.get(analysis_type, analysis_prompts['general'])}\n\nFile: {filepath}\n\nCode:\n```\n{content}\n```"
|
|
|
|
system_prompt = "You are a code review expert. Provide detailed, constructive analysis."
|
|
|
|
response = llm.call_ollama(prompt, system_prompt)
|
|
|
|
return f"📋 Analysis of {filepath}:\n\n{response}"
|
|
|
|
except FileNotFoundError:
|
|
return f"❌ File not found: {filepath}"
|
|
except Exception as e:
|
|
logger.error(f"File analysis failed: {e}")
|
|
return f"❌ Error analyzing file: {e}"
|
|
|
|
@app.tool("write_code_to_file")
|
|
async def write_code_to_file(
|
|
filepath: str,
|
|
task_description: str,
|
|
overwrite: bool = False
|
|
) -> str:
|
|
"""
|
|
LLMでコードを生成してファイルに書き込み
|
|
|
|
Args:
|
|
filepath: 書き込み先のファイルパス
|
|
task_description: コード生成のタスク説明
|
|
overwrite: 既存ファイルを上書きするか
|
|
"""
|
|
logger.info(f"Generating code for file: {filepath}")
|
|
|
|
try:
|
|
# 既存ファイルのチェック
|
|
if os.path.exists(filepath) and not overwrite:
|
|
return f"❌ File already exists: {filepath}. Use overwrite=true to replace."
|
|
|
|
# ファイル拡張子から言語を推定
|
|
ext = Path(filepath).suffix.lower()
|
|
language_map = {
|
|
'.py': 'Python',
|
|
'.js': 'JavaScript',
|
|
'.ts': 'TypeScript',
|
|
'.java': 'Java',
|
|
'.cpp': 'C++',
|
|
'.c': 'C',
|
|
'.rs': 'Rust',
|
|
'.go': 'Go'
|
|
}
|
|
language = language_map.get(ext, 'appropriate language')
|
|
|
|
# コード生成プロンプト
|
|
prompt = f"""Generate {language} code for the following task and save it to {filepath}:
|
|
|
|
Task: {task_description}
|
|
|
|
Requirements:
|
|
- Write complete, working code
|
|
- Include appropriate comments
|
|
- Follow best practices for {language}
|
|
- Make the code production-ready
|
|
|
|
Return ONLY the code that should be saved to the file, without any additional explanation or markdown formatting."""
|
|
|
|
system_prompt = f"You are an expert {language} developer. Generate clean, efficient, well-documented code."
|
|
|
|
# コード生成
|
|
generated_code = llm.call_ollama(prompt, system_prompt)
|
|
|
|
# ファイルに書き込み
|
|
os.makedirs(os.path.dirname(filepath), exist_ok=True)
|
|
with open(filepath, 'w', encoding='utf-8') as f:
|
|
f.write(generated_code)
|
|
|
|
return f"✅ Code generated and saved to {filepath}\n\nGenerated code:\n```{language.lower()}\n{generated_code}\n```"
|
|
|
|
except Exception as e:
|
|
logger.error(f"Code generation and file writing failed: {e}")
|
|
return f"❌ Error: {e}"
|
|
|
|
@app.tool("debug_with_llm")
|
|
async def debug_with_llm(
|
|
error_message: str,
|
|
code_context: str = "",
|
|
filepath: str = ""
|
|
) -> str:
|
|
"""
|
|
エラーメッセージとコードコンテキストでデバッグ支援
|
|
|
|
Args:
|
|
error_message: エラーメッセージ
|
|
code_context: エラーが発生したコードの部分
|
|
filepath: エラーが発生したファイル(オプション)
|
|
"""
|
|
logger.info("Debugging with LLM")
|
|
|
|
try:
|
|
# ファイルが指定されていれば読み込み
|
|
if filepath and os.path.exists(filepath):
|
|
with open(filepath, 'r', encoding='utf-8', errors='ignore') as f:
|
|
file_content = f.read()
|
|
code_context = f"Full file content:\n{file_content}"
|
|
|
|
prompt = f"""Help debug this error:
|
|
|
|
Error message: {error_message}
|
|
|
|
Code context:
|
|
{code_context}
|
|
|
|
Please:
|
|
1. Explain what's causing the error
|
|
2. Provide a specific solution
|
|
3. Show the corrected code if applicable
|
|
4. Suggest ways to prevent similar errors"""
|
|
|
|
system_prompt = "You are an expert debugger. Provide clear, actionable solutions to programming errors."
|
|
|
|
response = llm.call_ollama(prompt, system_prompt)
|
|
|
|
return f"🔧 Debug Analysis:\n\n{response}"
|
|
|
|
except Exception as e:
|
|
logger.error(f"Debugging failed: {e}")
|
|
return f"❌ Debug error: {e}"
|
|
|
|
@app.tool("explain_code")
|
|
async def explain_code(
|
|
code: str,
|
|
detail_level: str = "medium"
|
|
) -> str:
|
|
"""
|
|
コードの説明を生成
|
|
|
|
Args:
|
|
code: 説明するコード
|
|
detail_level: 説明の詳細レベル (basic, medium, detailed)
|
|
"""
|
|
logger.info("Explaining code with LLM")
|
|
|
|
try:
|
|
detail_prompts = {
|
|
"basic": "Provide a brief, high-level explanation of what this code does.",
|
|
"medium": "Explain this code in detail, including its purpose, how it works, and key components.",
|
|
"detailed": "Provide a comprehensive explanation including line-by-line analysis, design patterns used, and potential improvements."
|
|
}
|
|
|
|
prompt = f"{detail_prompts.get(detail_level, detail_prompts['medium'])}\n\nCode:\n```\n{code}\n```"
|
|
|
|
system_prompt = "You are a programming instructor. Explain code clearly and educationally."
|
|
|
|
response = llm.call_ollama(prompt, system_prompt)
|
|
|
|
return f"📚 Code Explanation:\n\n{response}"
|
|
|
|
except Exception as e:
|
|
logger.error(f"Code explanation failed: {e}")
|
|
return f"❌ Explanation error: {e}"
|
|
|
|
@app.tool("switch_model")
|
|
async def switch_model(model_name: str) -> str:
|
|
"""
|
|
使用するローカルLLMモデルを切り替え
|
|
|
|
Args:
|
|
model_name: 切り替え先のモデル名
|
|
"""
|
|
logger.info(f"Switching model to: {model_name}")
|
|
|
|
try:
|
|
# モデルの存在確認
|
|
response = requests.get(f"{llm.ollama_url}/api/tags")
|
|
if response.status_code == 200:
|
|
models = response.json().get("models", [])
|
|
available_models = [model["name"] for model in models]
|
|
|
|
if model_name in available_models:
|
|
llm.model = model_name
|
|
return f"✅ Model switched to: {model_name}"
|
|
else:
|
|
return f"❌ Model not found. Available models: {', '.join(available_models)}"
|
|
else:
|
|
return "❌ Cannot check available models"
|
|
|
|
except Exception as e:
|
|
logger.error(f"Model switching failed: {e}")
|
|
return f"❌ Error switching model: {e}"
|
|
|
|
async def main():
|
|
"""MCPサーバーを起動"""
|
|
logger.info("Starting Local LLM MCP Server...")
|
|
logger.info(f"Using model: {llm.model}")
|
|
|
|
# Ollamaの接続確認
|
|
try:
|
|
response = requests.get(f"{llm.ollama_url}/api/tags", timeout=5)
|
|
if response.status_code == 200:
|
|
logger.info("✅ Ollama connection successful")
|
|
else:
|
|
logger.warning("⚠️ Ollama connection issue")
|
|
except Exception as e:
|
|
logger.error(f"❌ Cannot connect to Ollama: {e}")
|
|
|
|
# サーバー起動
|
|
await app.run()
|
|
|
|
if __name__ == "__main__":
|
|
asyncio.run(main()) |