commit 1a09129395bd11f28a48751836cd7fa5dbac3b1c
Author: liusongtao <sonto.lau@gmail.com>
Date:   Sat Feb 28 16:21:35 2026 +0800

    first agent project

diff --git a/client/__init__.py b/client/__init__.py
new file mode 100644
index 0000000..912a555
--- /dev/null
+++ b/client/__init__.py
@@ -0,0 +1 @@
+"""__init__.py"""
diff --git a/client/__pycache__/__init__.cpython-310.pyc b/client/__pycache__/__init__.cpython-310.pyc
new file mode 100644
index 0000000..226faba
Binary files /dev/null and b/client/__pycache__/__init__.cpython-310.pyc differ
diff --git a/client/__pycache__/agent_client.cpython-310.pyc b/client/__pycache__/agent_client.cpython-310.pyc
new file mode 100644
index 0000000..5ab36a2
Binary files /dev/null and b/client/__pycache__/agent_client.cpython-310.pyc differ
diff --git a/client/agent_client.py b/client/agent_client.py
new file mode 100644
index 0000000..802f889
--- /dev/null
+++ b/client/agent_client.py
@@ -0,0 +1,158 @@
+"""客户端：用户交互 & 会话管理"""
+"""
+client/agent_client.py
+Agent 客户端：协调 LLM 引擎、MCP 服务器、记忆模块，驱动完整 Agent 执行流程
+"""
+
+from dataclasses import dataclass
+
+from llm.llm_engine import LLMEngine
+from mcp.mcp_protocol import MCPMethod, MCPRequest
+from mcp.mcp_server import MCPServer
+from memory.memory_store import MemoryStore
+from utils.logger import get_logger
+
+
+# ── 单轮执行结果 ───────────────────────────────────────────────
+@dataclass
+class AgentResponse:
+    """一次完整 Agent 调用的结果"""
+    user_input:   str
+    final_reply:  str
+    tool_used:    str | None = None
+    tool_output:  str | None = None
+    success:      bool       = True
+    error:        str | None = None
+
+
+# ── Agent 客户端 ───────────────────────────────────────────────
+class AgentClient:
+    """
+    Agent 客户端：实现完整的 ReAct 执行循环
+
+    执行流程 (5步):
+        1. [CLIENT]  接收用户输入，写入 Memory
+        2. [LLM]     分析意图，决策是否调用工具
+        3. [MCP]     构造 JSON-RPC 请求，发送给 MCP Server
+        4. [TOOL]    MCP Server 执行工具，返回结果
+        5. [LLM]     整合结果，生成最终回复，写入 Memory
+
+    使用示例:
+        client = AgentClient(llm=llm, mcp_server=mcp, memory=memory)
+        response = client.chat("帮我计算 100 * 200")
+        print(response.final_reply)
+    """
+
+    def __init__(
+        self,
+        llm:        LLMEngine,
+        mcp_server: MCPServer,
+        memory:     MemoryStore,
+    ):
+        self.llm        = llm
+        self.mcp_server = mcp_server
+        self.memory     = memory
+        self.logger     = get_logger("CLIENT")
+        self.logger.info("💻 Agent Client 初始化完成")
+
+    # ── 主入口 ──────────────────────────────────────────────────
+
+    def chat(self, user_input: str) -> AgentResponse:
+        """
+        处理一轮用户对话，执行完整 Agent 流程
+
+        Args:
+            user_input: 用户输入的自然语言文本
+
+        Returns:
+            AgentResponse 实例
+        """
+        self.logger.info(f"{'='*55}")
+        self.logger.info(f"📨 Step 1 [CLIENT] 收到用户输入: {user_input}")
+
+        # ── Step 1: 记录用户消息 ────────────────────────────────
+        self.memory.add_user_message(user_input)
+        context = self.memory.get_context_summary()
+
+        # ── Step 2: LLM 推理决策 ────────────────────────────────
+        self.logger.info("🧠 Step 2 [LLM] 开始推理，分析意图...")
+        tool_schemas = self.mcp_server.get_tool_schemas()
+        decision     = self.llm.think_and_decide(user_input, tool_schemas, context)
+
+        # ── 分支：是否需要工具 ──────────────────────────────────
+        if not decision.need_tool:
+            return self._handle_direct_reply(user_input, context)
+
+        return self._handle_tool_call(user_input, decision, context)
+
+    # ── 私有流程方法 ────────────────────────────────────────────
+
+    def _handle_direct_reply(self, user_input: str, context: str) -> AgentResponse:
+        """无需工具时直接生成回复"""
+        self.logger.info("💬 无需工具，直接生成回复")
+        reply = self.llm.generate_direct_reply(user_input, context)
+        self.memory.add_assistant_message(reply)
+        return AgentResponse(user_input=user_input, final_reply=reply)
+
+    def _handle_tool_call(
+        self,
+        user_input: str,
+        decision,
+        context:    str,
+    ) -> AgentResponse:
+        """执行工具调用的完整流程（Step 3 → 4 → 5）"""
+
+        # ── Step 3: 构造 MCP 请求 ───────────────────────────────
+        mcp_request: MCPRequest = decision.to_mcp_request()
+        self.logger.info(
+            f"📡 Step 3 [MCP] 发送工具调用请求\n"
+            f"    方法: {mcp_request.method}\n"
+            f"    工具: {decision.tool_name}\n"
+            f"    参数: {decision.arguments}\n"
+            f"    请求体: {mcp_request.to_dict()}"
+        )
+
+        # ── Step 4: MCP Server 执行工具 ─────────────────────────
+        self.logger.info(f"🔧 Step 4 [TOOL] MCP Server 执行工具 [{decision.tool_name}]...")
+        mcp_response = self.mcp_server.handle_request(mcp_request)
+
+        if not mcp_response.success:
+            error_msg = f"工具调用失败: {mcp_response.error}"
+            self.logger.error(f"❌ {error_msg}")
+            return AgentResponse(
+                user_input=user_input,
+                final_reply=f"抱歉，工具调用失败：{mcp_response.error.get('message')}",
+                tool_used=decision.tool_name,
+                success=False,
+                error=error_msg,
+            )
+
+        tool_output = mcp_response.content
+        self.logger.info(f"✅ 工具执行成功，输出: {tool_output[:80]}...")
+        self.memory.add_tool_result(decision.tool_name, tool_output)
+
+        # ── Step 5: LLM 整合结果，生成最终回复 ──────────────────
+        self.logger.info("✍️  Step 5 [LLM] 整合工具结果，生成最终回复...")
+        final_reply = self.llm.generate_final_reply(
+            user_input, decision.tool_name, tool_output, context
+        )
+        self.memory.add_assistant_message(final_reply)
+
+        self.logger.info(f"🎉 [CLIENT] 流程完成，回复已返回")
+        return AgentResponse(
+            user_input=user_input,
+            final_reply=final_reply,
+            tool_used=decision.tool_name,
+            tool_output=tool_output,
+        )
+
+    # ── 工具方法 ────────────────────────────────────────────────
+
+    def get_memory_stats(self) -> dict:
+        """获取当前记忆统计"""
+        return self.memory.stats()
+
+    def clear_session(self) -> None:
+        """清空当前会话"""
+        self.memory.clear_history()
+        self.logger.info("🗑 会话已清空")
\ No newline at end of file
diff --git a/llm/__init__.py b/llm/__init__.py
new file mode 100644
index 0000000..912a555
--- /dev/null
+++ b/llm/__init__.py
@@ -0,0 +1 @@
+"""__init__.py"""
diff --git a/llm/__pycache__/__init__.cpython-310.pyc b/llm/__pycache__/__init__.cpython-310.pyc
new file mode 100644
index 0000000..232cd1d
Binary files /dev/null and b/llm/__pycache__/__init__.cpython-310.pyc differ
diff --git a/llm/__pycache__/llm_engine.cpython-310.pyc b/llm/__pycache__/llm_engine.cpython-310.pyc
new file mode 100644
index 0000000..b5889d6
Binary files /dev/null and b/llm/__pycache__/llm_engine.cpython-310.pyc differ
diff --git a/llm/llm_engine.py b/llm/llm_engine.py
new file mode 100644
index 0000000..843fc85
--- /dev/null
+++ b/llm/llm_engine.py
@@ -0,0 +1,248 @@
+"""LLM 引擎：意图理解 & 工具决策"""
+"""
+llm/llm_engine.py
+LLM 引擎：负责意图理解、工具选择决策、最终回复生成
+生产环境可替换 _call_llm_api() 为真实 API 调用（OpenAI / Anthropic 等）
+"""
+
+import json
+import re
+from dataclasses import dataclass
+
+from mcp.mcp_protocol import MCPRequest, MCPMethod, ToolSchema
+from utils.logger import get_logger
+from openai import OpenAI
+
+
+# ── 工具调用决策结果 ───────────────────────────────────────────
+@dataclass
+class ToolDecision:
+    """LLM 决策是否调用工具及调用参数"""
+    need_tool:  bool
+    tool_name:  str        = ""
+    arguments:  dict       = None
+    reasoning:  str        = ""   # 推理过程说明
+
+    def __post_init__(self):
+        self.arguments = self.arguments or {}
+
+    def to_mcp_request(self) -> MCPRequest | None:
+        """将工具决策转换为 MCP 请求"""
+        if not self.need_tool:
+            return None
+        return MCPRequest(
+            method=MCPMethod.TOOLS_CALL,
+            params={"name": self.tool_name, "arguments": self.arguments},
+        )
+
+class MonicaClient:
+
+    BASE_URL = "https://openapi.monica.im/v1"
+    def __init__(self, api_key):
+        self.client = OpenAI(base_url=self.BASE_URL,
+                             api_key=api_key)
+    def create(self, model_name, tool_schemas, user_input) -> ToolDecision:
+        tools = [{
+                "name": s.name,
+                "description": s.description,
+                "parameters": s.parameters} for s in tool_schemas]
+        completion = self.client.chat.completions.create(
+            model=model_name,
+            functions=tools,
+            messages = [
+                {
+                     "role": "user",
+                     "content": [{
+                                    "type": "text",
+                                    "text": user_input
+                                }]
+                 }
+            ]
+        )
+        response = json.loads(completion.choices[0].message.content)
+        return ToolDecision(need_tool=response['need_tool'],
+                            tool_name=response['tool_name'],
+                            arguments=response['arguments'],
+                            reasoning=response['reasoning'])
+
+# ── LLM 引擎 ──────────────────────────────────────────────────
+class LLMEngine:
+    """
+    LLM 推理引擎（ReAct 模式）
+
+    执行流程:
+        1. 接收用户输入 + 工具列表
+        2. 分析意图，决策是否调用工具（think）
+        3. 若需要工具，生成 MCPRequest（act）
+        4. 接收工具结果，生成最终回复（observe）
+
+    生产环境替换:
+        将 _call_llm_api() 替换为真实 LLM API 调用即可，
+        其余流程控制逻辑保持不变。
+    """
+
+    API_KEY = "sk-AUmOuFI731Ty5Nob38jY26d8lydfDT-QkE2giqb0sCuPCAE2JH6zjLM4lZLpvL5WMYPOocaMe2FwVDmqM_9KimmKACjR"
+    def __init__(self, model_name: str = "claude-sonnet-4-6"):
+        self.model_name = model_name
+        self.logger     = get_logger("LLM")
+        self.logger.info(f"🧠 LLM 引擎初始化，模型: {model_name}")
+        self.client = MonicaClient(api_key=self.API_KEY)
+
+    # ── 核心推理流程 ────────────────────────────────────────────
+
+    def think_and_decide(
+        self,
+        user_input:   str,
+        tool_schemas: list[ToolSchema],
+        context:      str = "",
+    ) -> ToolDecision:
+        """
+        Step 1 & 2: 理解意图，决策工具调用（Think 阶段）
+
+        Args:
+            user_input:   用户输入文本
+            tool_schemas: 可用工具的 Schema 列表
+            context:      对话历史上下文摘要
+
+        Returns:
+            ToolDecision 实例
+        """
+        self.logger.info(f"💭 分析意图: {user_input[:50]}...")
+
+        # 构造 Prompt（生产环境发送给真实 LLM）
+        prompt = self._build_decision_prompt(user_input, tool_schemas, context)
+        self.logger.debug(f"📝 Prompt 已构造 ({len(prompt)} chars)")
+
+        # 调用 LLM（Demo 中使用规则模拟）
+        # decision = self._call_llm_api(user_input, tool_schemas)
+        decision = self._call_llm_api(prompt, tool_schemas)
+
+        self.logger.info(
+            f"🎯 决策结果: {'调用工具 [' + decision.tool_name + ']' if decision.need_tool else '直接回复'}"
+        )
+        self.logger.debug(f"💡 推理: {decision.reasoning}")
+        return decision
+
+    def generate_final_reply(
+        self,
+        user_input:   str,
+        tool_name:    str,
+        tool_output:  str,
+        context:      str = "",
+    ) -> str:
+        """
+        Step 5: 整合工具结果，生成最终自然语言回复（Observe 阶段）
+
+        Args:
+            user_input:  原始用户输入
+            tool_name:   被调用的工具名称
+            tool_output: 工具返回的原始输出
+            context:     对话历史上下文
+
+        Returns:
+            最终回复字符串
+        """
+        self.logger.info("✍️  整合工具结果，生成最终回复...")
+
+        # 生产环境：将 tool_output 注入 Prompt，调用 LLM 生成回复
+        reply = self._synthesize_reply(user_input, tool_name, tool_output)
+        self.logger.info(f"💬 回复已生成 ({len(reply)} chars)")
+        return reply
+
+    def generate_direct_reply(self, user_input: str, context: str = "") -> str:
+        """无需工具时直接生成回复"""
+        self.logger.info("💬 直接生成回复（无需工具）")
+        return f"[{self.model_name}] 您好！关于「{user_input}」，这是一个直接回复示例。\n（生产环境此处调用真实 LLM API）"
+
+    # ── Prompt 构造 ─────────────────────────────────────────────
+
+    def _build_decision_prompt(
+        self,
+        user_input:   str,
+        tool_schemas: list[ToolSchema],
+        context:      str,
+    ) -> str:
+        """构造工具决策 Prompt（ReAct 格式）"""
+        tools_desc = "\n".join(
+            f"- {s.name}: {s.description}" for s in tool_schemas
+        )
+        return (
+            f"你是一个智能助手，请分析用户输入并决定是否需要调用工具。\n\n"
+            f"## 可用工具\n{tools_desc}\n\n"
+            f"## 对话历史\n{context or '（无）'}\n\n"
+            f"## 用户输入\n{user_input}\n\n"
+            f"## 指令\n"
+            f"以 JSON 格式回复：\n"
+            f'{{"need_tool": true/false, "tool_name": "...", "arguments": {{...}}, "reasoning": "..."}}'
+        )
+
+    # ── 模拟 LLM API（Demo 用规则引擎替代）────────────────────
+
+    def _call_llm_api(self, user_input: str, tool_schemas: list[ToolSchema]) -> ToolDecision:
+        """
+        模拟 LLM API 调用（Demo 版本使用关键词规则）
+
+        生产环境替换示例:
+            import anthropic
+            client = anthropic.Anthropic()
+            response = client.messages.create(
+                model=self.model_name,
+                tools=[s.to_dict() for s in tool_schemas],
+                messages=[{"role": "user", "content": user_input}]
+            )
+            # 解析 response.content 中的 tool_use block
+        """
+        return self.client.create(self.model_name, user_input=user_input, tool_schemas=tool_schemas)
+
+        text = user_input.lower()
+
+        # 规则匹配：计算器
+        calc_pattern = re.search(r"[\d\s\+\-\*\/\(\)\^]+[=？?]?", user_input)
+        if any(kw in text for kw in ["计算", "等于", "多少", "×", "÷"]) and calc_pattern:
+            expr = re.sub(r"[^0-9+\-*/().**]", "", user_input.replace("×","*").replace("÷","/"))
+            return ToolDecision(
+                need_tool=True, tool_name="calculator",
+                arguments={"expression": expr or "1+1"},
+                reasoning="用户请求数学计算，调用 calculator 工具",
+            )
+
+        # 规则匹配：搜索
+        if any(kw in text for kw in ["搜索", "查询", "天气", "新闻", "查一下", "search"]):
+            return ToolDecision(
+                need_tool=True, tool_name="web_search",
+                arguments={"query": user_input, "max_results": 3},
+                reasoning="用户需要实时信息，调用 web_search 工具",
+            )
+
+        # 规则匹配：文件读取
+        if any(kw in text for kw in ["文件", "读取", "file", "config", "json", "txt"]):
+            filename = re.search(r"[\w\-\.]+\.\w+", user_input)
+            return ToolDecision(
+                need_tool=True, tool_name="file_reader",
+                arguments={"path": filename.group() if filename else "config.json"},
+                reasoning="用户请求读取文件，调用 file_reader 工具",
+            )
+
+        # 规则匹配：代码执行
+        if any(kw in text for kw in ["执行", "运行", "代码", "python", "print", "code"]):
+            code_match = re.search(r'[`\'"](.+?)[`\'"]', user_input)
+            code = code_match.group(1) if code_match else 'print("Hello, Agent!")'
+            return ToolDecision(
+                need_tool=True, tool_name="code_executor",
+                arguments={"code": code, "timeout": 5},
+                reasoning="用户请求执行代码，调用 code_executor 工具",
+            )
+
+        # 默认：直接回复
+        return ToolDecision(
+            need_tool=False,
+            reasoning="问题可直接回答，无需工具",
+        )
+
+    def _synthesize_reply(self, user_input: str, tool_name: str, tool_output: str) -> str:
+        """基于工具输出合成最终回复（Demo 版本）"""
+        return (
+            f"✅ 已通过 [{tool_name}] 工具处理您的请求。\n\n"
+            f"**执行结果:**\n{tool_output}\n\n"
+            f"---\n*由 {self.model_name} 生成 · 工具: {tool_name}*"
+        )
\ No newline at end of file
diff --git a/logs/agent.log b/logs/agent.log
new file mode 100644
index 0000000..00ba484
--- /dev/null
+++ b/logs/agent.log
@@ -0,0 +1,210 @@
+[2026-02-28 13:16:51,269] [agent.SYSTEM] INFO: 🔧 开始组装 Agent 系统...
+[2026-02-28 13:16:51,270] [agent.MCP] INFO: 🚀 MCP Server [DemoMCPServer] 启动
+[2026-02-28 13:16:51,271] [agent.MCP] INFO: 📌 注册工具: [calculator] — 计算数学表达式，支持加减乘除、幂运算、括号等
+[2026-02-28 13:16:51,271] [agent.MCP] INFO: 📌 注册工具: [web_search] — 在互联网上搜索信息，返回相关网页摘要
+[2026-02-28 13:16:51,271] [agent.MCP] INFO: 📌 注册工具: [file_reader] — 读取本地文件内容，仅限 workspace/ 目录下的文件
+[2026-02-28 13:16:51,271] [agent.MCP] INFO: 📌 注册工具: [code_executor] — 在沙箱环境中执行 Python 代码片段，返回标准输出
+[2026-02-28 13:16:51,272] [agent.LLM] INFO: 🧠 LLM 引擎初始化，模型: claude-sonnet-4-6
+[2026-02-28 13:16:51,273] [agent.MEMORY] INFO: 💾 Memory 初始化，最大历史: 20 条
+[2026-02-28 13:16:51,273] [agent.CLIENT] INFO: 💻 Agent Client 初始化完成
+[2026-02-28 13:16:51,273] [agent.SYSTEM] INFO: ✅ Agent 组装完成，已注册工具: ['calculator', 'web_search', 'file_reader', 'code_executor']
+[2026-02-28 13:59:08,843] [agent.SYSTEM] INFO: 🔧 开始组装 Agent 系统...
+[2026-02-28 13:59:08,844] [agent.MCP] INFO: 🚀 MCP Server [DemoMCPServer] 启动
+[2026-02-28 13:59:08,845] [agent.MCP] INFO: 📌 注册工具: [calculator] — 计算数学表达式，支持加减乘除、幂运算、括号等
+[2026-02-28 13:59:08,845] [agent.MCP] INFO: 📌 注册工具: [web_search] — 在互联网上搜索信息，返回相关网页摘要
+[2026-02-28 13:59:08,845] [agent.MCP] INFO: 📌 注册工具: [file_reader] — 读取本地文件内容，仅限 workspace/ 目录下的文件
+[2026-02-28 13:59:08,845] [agent.MCP] INFO: 📌 注册工具: [code_executor] — 在沙箱环境中执行 Python 代码片段，返回标准输出
+[2026-02-28 13:59:08,846] [agent.LLM] INFO: 🧠 LLM 引擎初始化，模型: claude-sonnet-4-6
+[2026-02-28 13:59:08,847] [agent.MEMORY] INFO: 💾 Memory 初始化，最大历史: 20 条
+[2026-02-28 13:59:08,848] [agent.CLIENT] INFO: 💻 Agent Client 初始化完成
+[2026-02-28 13:59:08,848] [agent.SYSTEM] INFO: ✅ Agent 组装完成，已注册工具: ['calculator', 'web_search', 'file_reader', 'code_executor']
+[2026-02-28 13:59:40,111] [agent.CLIENT] INFO: =======================================================
+[2026-02-28 13:59:41,400] [agent.CLIENT] INFO: 📨 Step 1 [CLIENT] 收到用户输入: 1加1等于多少
+[2026-02-28 14:00:07,071] [agent.MEMORY] DEBUG: 💬 [USER] 1加1等于多少...
+[2026-02-28 14:00:51,574] [agent.CLIENT] INFO: 🧠 Step 2 [LLM] 开始推理，分析意图...
+[2026-02-28 14:02:14,371] [agent.LLM] INFO: 💭 分析意图: 1加1等于多少...
+[2026-02-28 14:04:03,060] [agent.LLM] DEBUG: 📝 Prompt 已构造 (344 chars)
+[2026-02-28 14:15:21,435] [agent.SYSTEM] INFO: 🔧 开始组装 Agent 系统...
+[2026-02-28 14:15:21,437] [agent.MCP] INFO: 🚀 MCP Server [DemoMCPServer] 启动
+[2026-02-28 14:15:21,437] [agent.MCP] INFO: 📌 注册工具: [calculator] — 计算数学表达式，支持加减乘除、幂运算、括号等
+[2026-02-28 14:15:21,438] [agent.MCP] INFO: 📌 注册工具: [web_search] — 在互联网上搜索信息，返回相关网页摘要
+[2026-02-28 14:15:21,438] [agent.MCP] INFO: 📌 注册工具: [file_reader] — 读取本地文件内容，仅限 workspace/ 目录下的文件
+[2026-02-28 14:15:21,438] [agent.MCP] INFO: 📌 注册工具: [code_executor] — 在沙箱环境中执行 Python 代码片段，返回标准输出
+[2026-02-28 14:15:21,439] [agent.LLM] INFO: 🧠 LLM 引擎初始化，模型: claude-sonnet-4-6
+[2026-02-28 14:15:21,730] [agent.MEMORY] INFO: 💾 Memory 初始化，最大历史: 20 条
+[2026-02-28 14:15:21,731] [agent.CLIENT] INFO: 💻 Agent Client 初始化完成
+[2026-02-28 14:15:21,731] [agent.SYSTEM] INFO: ✅ Agent 组装完成，已注册工具: ['calculator', 'web_search', 'file_reader', 'code_executor']
+[2026-02-28 14:15:37,946] [agent.CLIENT] INFO: =======================================================
+[2026-02-28 14:15:38,634] [agent.CLIENT] INFO: 📨 Step 1 [CLIENT] 收到用户输入: 1+1等于多少
+[2026-02-28 14:15:39,212] [agent.MEMORY] DEBUG: 💬 [USER] 1+1等于多少...
+[2026-02-28 14:15:40,563] [agent.CLIENT] INFO: 🧠 Step 2 [LLM] 开始推理，分析意图...
+[2026-02-28 14:15:47,714] [agent.LLM] INFO: 💭 分析意图: 1+1等于多少...
+[2026-02-28 14:15:50,157] [agent.LLM] DEBUG: 📝 Prompt 已构造 (344 chars)
+[2026-02-28 14:17:52,834] [agent.SYSTEM] INFO: 🔧 开始组装 Agent 系统...
+[2026-02-28 14:17:52,836] [agent.MCP] INFO: 🚀 MCP Server [DemoMCPServer] 启动
+[2026-02-28 14:17:52,837] [agent.MCP] INFO: 📌 注册工具: [calculator] — 计算数学表达式，支持加减乘除、幂运算、括号等
+[2026-02-28 14:17:52,837] [agent.MCP] INFO: 📌 注册工具: [web_search] — 在互联网上搜索信息，返回相关网页摘要
+[2026-02-28 14:17:52,837] [agent.MCP] INFO: 📌 注册工具: [file_reader] — 读取本地文件内容，仅限 workspace/ 目录下的文件
+[2026-02-28 14:17:52,838] [agent.MCP] INFO: 📌 注册工具: [code_executor] — 在沙箱环境中执行 Python 代码片段，返回标准输出
+[2026-02-28 14:17:52,839] [agent.LLM] INFO: 🧠 LLM 引擎初始化，模型: claude-sonnet-4-6
+[2026-02-28 14:17:53,009] [agent.MEMORY] INFO: 💾 Memory 初始化，最大历史: 20 条
+[2026-02-28 14:17:53,010] [agent.CLIENT] INFO: 💻 Agent Client 初始化完成
+[2026-02-28 14:17:53,010] [agent.SYSTEM] INFO: ✅ Agent 组装完成，已注册工具: ['calculator', 'web_search', 'file_reader', 'code_executor']
+[2026-02-28 14:18:02,052] [agent.CLIENT] INFO: =======================================================
+[2026-02-28 14:18:02,614] [agent.CLIENT] INFO: 📨 Step 1 [CLIENT] 收到用户输入: 1加1等于多少
+[2026-02-28 14:18:03,083] [agent.MEMORY] DEBUG: 💬 [USER] 1加1等于多少...
+[2026-02-28 14:18:04,093] [agent.CLIENT] INFO: 🧠 Step 2 [LLM] 开始推理，分析意图...
+[2026-02-28 14:18:07,998] [agent.LLM] INFO: 💭 分析意图: 1加1等于多少...
+[2026-02-28 14:18:09,082] [agent.LLM] DEBUG: 📝 Prompt 已构造 (344 chars)
+[2026-02-28 14:19:28,781] [agent.SYSTEM] INFO: 🔧 开始组装 Agent 系统...
+[2026-02-28 14:19:28,781] [agent.MCP] INFO: 🚀 MCP Server [DemoMCPServer] 启动
+[2026-02-28 14:19:28,782] [agent.MCP] INFO: 📌 注册工具: [calculator] — 计算数学表达式，支持加减乘除、幂运算、括号等
+[2026-02-28 14:19:28,782] [agent.MCP] INFO: 📌 注册工具: [web_search] — 在互联网上搜索信息，返回相关网页摘要
+[2026-02-28 14:19:28,783] [agent.MCP] INFO: 📌 注册工具: [file_reader] — 读取本地文件内容，仅限 workspace/ 目录下的文件
+[2026-02-28 14:19:28,783] [agent.MCP] INFO: 📌 注册工具: [code_executor] — 在沙箱环境中执行 Python 代码片段，返回标准输出
+[2026-02-28 14:19:28,783] [agent.LLM] INFO: 🧠 LLM 引擎初始化，模型: gpt-4o
+[2026-02-28 14:19:28,912] [agent.MEMORY] INFO: 💾 Memory 初始化，最大历史: 20 条
+[2026-02-28 14:19:28,913] [agent.CLIENT] INFO: 💻 Agent Client 初始化完成
+[2026-02-28 14:19:28,913] [agent.SYSTEM] INFO: ✅ Agent 组装完成，已注册工具: ['calculator', 'web_search', 'file_reader', 'code_executor']
+[2026-02-28 14:19:38,773] [agent.CLIENT] INFO: =======================================================
+[2026-02-28 14:19:38,773] [agent.CLIENT] INFO: 📨 Step 1 [CLIENT] 收到用户输入: 1加1等于多少
+[2026-02-28 14:19:38,774] [agent.MEMORY] DEBUG: 💬 [USER] 1加1等于多少...
+[2026-02-28 14:19:38,774] [agent.CLIENT] INFO: 🧠 Step 2 [LLM] 开始推理，分析意图...
+[2026-02-28 14:19:38,774] [agent.LLM] INFO: 💭 分析意图: 1加1等于多少...
+[2026-02-28 14:19:38,774] [agent.LLM] DEBUG: 📝 Prompt 已构造 (344 chars)
+[2026-02-28 14:24:25,989] [agent.SYSTEM] INFO: 🔧 开始组装 Agent 系统...
+[2026-02-28 14:24:25,990] [agent.MCP] INFO: 🚀 MCP Server [DemoMCPServer] 启动
+[2026-02-28 14:24:25,991] [agent.MCP] INFO: 📌 注册工具: [calculator] — 计算数学表达式，支持加减乘除、幂运算、括号等
+[2026-02-28 14:24:25,991] [agent.MCP] INFO: 📌 注册工具: [web_search] — 在互联网上搜索信息，返回相关网页摘要
+[2026-02-28 14:24:25,991] [agent.MCP] INFO: 📌 注册工具: [file_reader] — 读取本地文件内容，仅限 workspace/ 目录下的文件
+[2026-02-28 14:24:25,991] [agent.MCP] INFO: 📌 注册工具: [code_executor] — 在沙箱环境中执行 Python 代码片段，返回标准输出
+[2026-02-28 14:24:25,992] [agent.LLM] INFO: 🧠 LLM 引擎初始化，模型: gpt-4o
+[2026-02-28 14:24:26,120] [agent.MEMORY] INFO: 💾 Memory 初始化，最大历史: 20 条
+[2026-02-28 14:24:26,120] [agent.CLIENT] INFO: 💻 Agent Client 初始化完成
+[2026-02-28 14:24:26,121] [agent.SYSTEM] INFO: ✅ Agent 组装完成，已注册工具: ['calculator', 'web_search', 'file_reader', 'code_executor']
+[2026-02-28 14:24:47,520] [agent.CLIENT] INFO: =======================================================
+[2026-02-28 14:24:47,520] [agent.CLIENT] INFO: 📨 Step 1 [CLIENT] 收到用户输入: 1加1等于多少
+[2026-02-28 14:24:47,521] [agent.MEMORY] DEBUG: 💬 [USER] 1加1等于多少...
+[2026-02-28 14:24:47,521] [agent.CLIENT] INFO: 🧠 Step 2 [LLM] 开始推理，分析意图...
+[2026-02-28 14:24:47,521] [agent.LLM] INFO: 💭 分析意图: 1加1等于多少...
+[2026-02-28 14:24:47,521] [agent.LLM] DEBUG: 📝 Prompt 已构造 (344 chars)
+[2026-02-28 14:26:41,628] [agent.SYSTEM] INFO: 🔧 开始组装 Agent 系统...
+[2026-02-28 14:26:41,629] [agent.MCP] INFO: 🚀 MCP Server [DemoMCPServer] 启动
+[2026-02-28 14:26:41,630] [agent.MCP] INFO: 📌 注册工具: [calculator] — 计算数学表达式，支持加减乘除、幂运算、括号等
+[2026-02-28 14:26:41,630] [agent.MCP] INFO: 📌 注册工具: [web_search] — 在互联网上搜索信息，返回相关网页摘要
+[2026-02-28 14:26:41,631] [agent.MCP] INFO: 📌 注册工具: [file_reader] — 读取本地文件内容，仅限 workspace/ 目录下的文件
+[2026-02-28 14:26:41,631] [agent.MCP] INFO: 📌 注册工具: [code_executor] — 在沙箱环境中执行 Python 代码片段，返回标准输出
+[2026-02-28 14:26:41,632] [agent.LLM] INFO: 🧠 LLM 引擎初始化，模型: gpt-4o
+[2026-02-28 14:26:41,764] [agent.MEMORY] INFO: 💾 Memory 初始化，最大历史: 20 条
+[2026-02-28 14:26:41,764] [agent.CLIENT] INFO: 💻 Agent Client 初始化完成
+[2026-02-28 14:26:41,765] [agent.SYSTEM] INFO: ✅ Agent 组装完成，已注册工具: ['calculator', 'web_search', 'file_reader', 'code_executor']
+[2026-02-28 14:26:52,266] [agent.CLIENT] INFO: =======================================================
+[2026-02-28 14:26:52,266] [agent.CLIENT] INFO: 📨 Step 1 [CLIENT] 收到用户输入: 1加1等于多少
+[2026-02-28 14:26:52,267] [agent.MEMORY] DEBUG: 💬 [USER] 1加1等于多少...
+[2026-02-28 14:26:52,267] [agent.CLIENT] INFO: 🧠 Step 2 [LLM] 开始推理，分析意图...
+[2026-02-28 14:26:52,267] [agent.LLM] INFO: 💭 分析意图: 1加1等于多少...
+[2026-02-28 14:26:52,268] [agent.LLM] DEBUG: 📝 Prompt 已构造 (344 chars)
+[2026-02-28 14:29:31,672] [agent.SYSTEM] INFO: 🔧 开始组装 Agent 系统...
+[2026-02-28 14:29:31,673] [agent.MCP] INFO: 🚀 MCP Server [DemoMCPServer] 启动
+[2026-02-28 14:29:31,674] [agent.MCP] INFO: 📌 注册工具: [calculator] — 计算数学表达式，支持加减乘除、幂运算、括号等
+[2026-02-28 14:29:31,674] [agent.MCP] INFO: 📌 注册工具: [web_search] — 在互联网上搜索信息，返回相关网页摘要
+[2026-02-28 14:29:31,674] [agent.MCP] INFO: 📌 注册工具: [file_reader] — 读取本地文件内容，仅限 workspace/ 目录下的文件
+[2026-02-28 14:29:31,675] [agent.MCP] INFO: 📌 注册工具: [code_executor] — 在沙箱环境中执行 Python 代码片段，返回标准输出
+[2026-02-28 14:29:31,675] [agent.LLM] INFO: 🧠 LLM 引擎初始化，模型: gpt-4o
+[2026-02-28 14:29:31,809] [agent.MEMORY] INFO: 💾 Memory 初始化，最大历史: 20 条
+[2026-02-28 14:29:31,810] [agent.CLIENT] INFO: 💻 Agent Client 初始化完成
+[2026-02-28 14:29:31,811] [agent.SYSTEM] INFO: ✅ Agent 组装完成，已注册工具: ['calculator', 'web_search', 'file_reader', 'code_executor']
+[2026-02-28 14:29:43,546] [agent.CLIENT] INFO: =======================================================
+[2026-02-28 14:29:43,546] [agent.CLIENT] INFO: 📨 Step 1 [CLIENT] 收到用户输入: 1加1等于多少
+[2026-02-28 14:29:43,546] [agent.MEMORY] DEBUG: 💬 [USER] 1加1等于多少...
+[2026-02-28 14:29:43,547] [agent.CLIENT] INFO: 🧠 Step 2 [LLM] 开始推理，分析意图...
+[2026-02-28 14:29:43,547] [agent.LLM] INFO: 💭 分析意图: 1加1等于多少...
+[2026-02-28 14:29:43,547] [agent.LLM] DEBUG: 📝 Prompt 已构造 (344 chars)
+[2026-02-28 14:32:01,347] [agent.SYSTEM] INFO: 🔧 开始组装 Agent 系统...
+[2026-02-28 14:32:01,348] [agent.MCP] INFO: 🚀 MCP Server [DemoMCPServer] 启动
+[2026-02-28 14:32:01,349] [agent.MCP] INFO: 📌 注册工具: [calculator] — 计算数学表达式，支持加减乘除、幂运算、括号等
+[2026-02-28 14:32:01,349] [agent.MCP] INFO: 📌 注册工具: [web_search] — 在互联网上搜索信息，返回相关网页摘要
+[2026-02-28 14:32:01,349] [agent.MCP] INFO: 📌 注册工具: [file_reader] — 读取本地文件内容，仅限 workspace/ 目录下的文件
+[2026-02-28 14:32:01,350] [agent.MCP] INFO: 📌 注册工具: [code_executor] — 在沙箱环境中执行 Python 代码片段，返回标准输出
+[2026-02-28 14:32:01,351] [agent.LLM] INFO: 🧠 LLM 引擎初始化，模型: gpt-4o
+[2026-02-28 14:32:01,477] [agent.MEMORY] INFO: 💾 Memory 初始化，最大历史: 20 条
+[2026-02-28 14:32:01,477] [agent.CLIENT] INFO: 💻 Agent Client 初始化完成
+[2026-02-28 14:32:01,478] [agent.SYSTEM] INFO: ✅ Agent 组装完成，已注册工具: ['calculator', 'web_search', 'file_reader', 'code_executor']
+[2026-02-28 14:32:08,959] [agent.CLIENT] INFO: =======================================================
+[2026-02-28 14:32:08,959] [agent.CLIENT] INFO: 📨 Step 1 [CLIENT] 收到用户输入: 1加1等于多少
+[2026-02-28 14:32:08,959] [agent.MEMORY] DEBUG: 💬 [USER] 1加1等于多少...
+[2026-02-28 14:32:08,960] [agent.CLIENT] INFO: 🧠 Step 2 [LLM] 开始推理，分析意图...
+[2026-02-28 14:32:08,960] [agent.LLM] INFO: 💭 分析意图: 1加1等于多少...
+[2026-02-28 14:32:08,960] [agent.LLM] DEBUG: 📝 Prompt 已构造 (344 chars)
+[2026-02-28 14:34:04,981] [agent.SYSTEM] INFO: 🔧 开始组装 Agent 系统...
+[2026-02-28 14:34:04,982] [agent.MCP] INFO: 🚀 MCP Server [DemoMCPServer] 启动
+[2026-02-28 14:34:04,983] [agent.MCP] INFO: 📌 注册工具: [calculator] — 计算数学表达式，支持加减乘除、幂运算、括号等
+[2026-02-28 14:34:04,984] [agent.MCP] INFO: 📌 注册工具: [web_search] — 在互联网上搜索信息，返回相关网页摘要
+[2026-02-28 14:34:04,984] [agent.MCP] INFO: 📌 注册工具: [file_reader] — 读取本地文件内容，仅限 workspace/ 目录下的文件
+[2026-02-28 14:34:04,985] [agent.MCP] INFO: 📌 注册工具: [code_executor] — 在沙箱环境中执行 Python 代码片段，返回标准输出
+[2026-02-28 14:34:04,987] [agent.LLM] INFO: 🧠 LLM 引擎初始化，模型: gpt-4o
+[2026-02-28 14:34:05,137] [agent.MEMORY] INFO: 💾 Memory 初始化，最大历史: 20 条
+[2026-02-28 14:34:05,140] [agent.CLIENT] INFO: 💻 Agent Client 初始化完成
+[2026-02-28 14:34:05,141] [agent.SYSTEM] INFO: ✅ Agent 组装完成，已注册工具: ['calculator', 'web_search', 'file_reader', 'code_executor']
+[2026-02-28 14:34:14,908] [agent.CLIENT] INFO: =======================================================
+[2026-02-28 14:34:14,908] [agent.CLIENT] INFO: 📨 Step 1 [CLIENT] 收到用户输入: 1加1等于多少
+[2026-02-28 14:34:14,908] [agent.MEMORY] DEBUG: 💬 [USER] 1加1等于多少...
+[2026-02-28 14:34:14,908] [agent.CLIENT] INFO: 🧠 Step 2 [LLM] 开始推理，分析意图...
+[2026-02-28 14:34:14,909] [agent.LLM] INFO: 💭 分析意图: 1加1等于多少...
+[2026-02-28 14:34:14,909] [agent.LLM] DEBUG: 📝 Prompt 已构造 (344 chars)
+[2026-02-28 14:35:39,204] [agent.SYSTEM] INFO: 🔧 开始组装 Agent 系统...
+[2026-02-28 14:35:39,205] [agent.MCP] INFO: 🚀 MCP Server [DemoMCPServer] 启动
+[2026-02-28 14:35:39,205] [agent.MCP] INFO: 📌 注册工具: [calculator] — 计算数学表达式，支持加减乘除、幂运算、括号等
+[2026-02-28 14:35:39,205] [agent.MCP] INFO: 📌 注册工具: [web_search] — 在互联网上搜索信息，返回相关网页摘要
+[2026-02-28 14:35:39,206] [agent.MCP] INFO: 📌 注册工具: [file_reader] — 读取本地文件内容，仅限 workspace/ 目录下的文件
+[2026-02-28 14:35:39,206] [agent.MCP] INFO: 📌 注册工具: [code_executor] — 在沙箱环境中执行 Python 代码片段，返回标准输出
+[2026-02-28 14:35:39,207] [agent.LLM] INFO: 🧠 LLM 引擎初始化，模型: gpt-4o
+[2026-02-28 14:35:39,333] [agent.MEMORY] INFO: 💾 Memory 初始化，最大历史: 20 条
+[2026-02-28 14:35:39,334] [agent.CLIENT] INFO: 💻 Agent Client 初始化完成
+[2026-02-28 14:35:39,334] [agent.SYSTEM] INFO: ✅ Agent 组装完成，已注册工具: ['calculator', 'web_search', 'file_reader', 'code_executor']
+[2026-02-28 14:35:51,583] [agent.CLIENT] INFO: =======================================================
+[2026-02-28 14:35:51,583] [agent.CLIENT] INFO: 📨 Step 1 [CLIENT] 收到用户输入: 1加1等于多少
+[2026-02-28 14:35:51,584] [agent.MEMORY] DEBUG: 💬 [USER] 1加1等于多少...
+[2026-02-28 14:35:51,584] [agent.CLIENT] INFO: 🧠 Step 2 [LLM] 开始推理，分析意图...
+[2026-02-28 14:35:51,584] [agent.LLM] INFO: 💭 分析意图: 1加1等于多少...
+[2026-02-28 14:35:51,585] [agent.LLM] DEBUG: 📝 Prompt 已构造 (344 chars)
+[2026-02-28 14:43:51,069] [agent.SYSTEM] INFO: 🔧 开始组装 Agent 系统...
+[2026-02-28 14:43:51,069] [agent.MCP] INFO: 🚀 MCP Server [DemoMCPServer] 启动
+[2026-02-28 14:43:51,070] [agent.MCP] INFO: 📌 注册工具: [calculator] — 计算数学表达式，支持加减乘除、幂运算、括号等
+[2026-02-28 14:43:51,070] [agent.MCP] INFO: 📌 注册工具: [web_search] — 在互联网上搜索信息，返回相关网页摘要
+[2026-02-28 14:43:51,071] [agent.MCP] INFO: 📌 注册工具: [file_reader] — 读取本地文件内容，仅限 workspace/ 目录下的文件
+[2026-02-28 14:43:51,071] [agent.MCP] INFO: 📌 注册工具: [code_executor] — 在沙箱环境中执行 Python 代码片段，返回标准输出
+[2026-02-28 14:43:51,071] [agent.LLM] INFO: 🧠 LLM 引擎初始化，模型: gpt-4o
+[2026-02-28 14:43:51,198] [agent.MEMORY] INFO: 💾 Memory 初始化，最大历史: 20 条
+[2026-02-28 14:43:51,198] [agent.CLIENT] INFO: 💻 Agent Client 初始化完成
+[2026-02-28 14:43:51,199] [agent.SYSTEM] INFO: ✅ Agent 组装完成，已注册工具: ['calculator', 'web_search', 'file_reader', 'code_executor']
+[2026-02-28 14:44:04,009] [agent.CLIENT] INFO: =======================================================
+[2026-02-28 14:44:04,009] [agent.CLIENT] INFO: 📨 Step 1 [CLIENT] 收到用户输入: 1加1等于多少
+[2026-02-28 14:44:04,009] [agent.MEMORY] DEBUG: 💬 [USER] 1加1等于多少...
+[2026-02-28 14:44:04,010] [agent.CLIENT] INFO: 🧠 Step 2 [LLM] 开始推理，分析意图...
+[2026-02-28 14:44:04,010] [agent.LLM] INFO: 💭 分析意图: 1加1等于多少...
+[2026-02-28 14:44:04,010] [agent.LLM] DEBUG: 📝 Prompt 已构造 (344 chars)
+[2026-02-28 14:44:10,159] [agent.LLM] INFO: 🎯 决策结果: 调用工具 [calculator]
+[2026-02-28 14:44:10,159] [agent.LLM] DEBUG: 💡 推理: 用户询问数学计算问题，直接调用计算工具得到1+1的结果是最合适的。
+[2026-02-28 14:44:21,419] [agent.CLIENT] INFO: 📡 Step 3 [MCP] 发送工具调用请求
+    方法: tools/call
+    工具: calculator
+    参数: {'expression': '1+1'}
+    请求体: {'jsonrpc': '2.0', 'id': 'e4d8732a', 'method': 'tools/call', 'params': {'name': 'calculator', 'arguments': {'expression': '1+1'}}}
+[2026-02-28 14:44:36,340] [agent.CLIENT] INFO: 🔧 Step 4 [TOOL] MCP Server 执行工具 [calculator]...
+[2026-02-28 14:44:40,181] [agent.MCP] INFO: 📨 收到请求 id=e4d8732a method=tools/call
+[2026-02-28 14:46:24,701] [agent.TOOL] INFO: ▶ 执行工具 [calculator]，参数: {'expression': '1+1'}
+[2026-02-28 14:46:50,239] [agent.TOOL] INFO: ✅ 工具 [calculator] 执行成功
+[2026-02-28 16:05:27,152] [agent.CLIENT] INFO: ✅ 工具执行成功，输出: 1+1 = 2...
+[2026-02-28 16:05:46,521] [agent.MEMORY] DEBUG: 💬 [TOOL] 1+1 = 2...
+[2026-02-28 16:05:49,469] [agent.CLIENT] INFO: ✍️  Step 5 [LLM] 整合工具结果，生成最终回复...
+[2026-02-28 16:05:53,270] [agent.LLM] INFO: ✍️  整合工具结果，生成最终回复...
+[2026-02-28 16:05:53,271] [agent.LLM] INFO: 💬 回复已生成 (83 chars)
+[2026-02-28 16:06:03,400] [agent.MEMORY] DEBUG: 💬 [ASSISTANT] ✅ 已通过 [calculator] 工具处理您的请求。
+
+**执行结果:**
+1+1 = 2
+
+---
+*由 gpt-...
+[2026-02-28 16:06:05,167] [agent.CLIENT] INFO: 🎉 [CLIENT] 流程完成，回复已返回
diff --git a/main.py b/main.py
new file mode 100644
index 0000000..48a7a60
--- /dev/null
+++ b/main.py
@@ -0,0 +1,160 @@
+"""程序入口"""
+"""
+main.py
+智能体 Demo 程序入口
+组装所有模块，启动交互式对话循环
+"""
+
+import sys
+
+# ── 导入各模块 ─────────────────────────────────────────────────
+from client.agent_client import AgentClient
+from llm.llm_engine import LLMEngine
+from mcp.mcp_server import MCPServer
+from memory.memory_store import MemoryStore
+from tools.calculator import CalculatorTool
+from tools.code_executor import CodeExecutorTool
+from tools.file_reader import FileReaderTool
+from tools.web_search import WebSearchTool
+from utils.logger import get_logger
+
+logger = get_logger("SYSTEM")
+
+
+# ── 系统组装 ───────────────────────────────────────────────────
+def build_agent() -> AgentClient:
+    """
+    工厂函数：组装并返回完整的 Agent 实例
+
+    组装顺序:
+        1. 初始化 MCP Server，注册所有工具
+        2. 初始化 LLM 引擎
+        3. 初始化 Memory 模块
+        4. 组装 AgentClient
+    """
+    logger.info("🔧 开始组装 Agent 系统...")
+
+    # 1. MCP Server：注册所有工具
+    mcp_server = MCPServer(server_name="DemoMCPServer")
+    mcp_server.register_tools(
+        CalculatorTool,
+        WebSearchTool,
+        FileReaderTool,
+        CodeExecutorTool,
+    )
+
+    # 2. LLM 引擎
+    llm = LLMEngine(model_name="gpt-4o")
+
+    # 3. 记忆模块
+    memory = MemoryStore(max_history=20)
+
+    # 4. 组装客户端
+    client = AgentClient(llm=llm, mcp_server=mcp_server, memory=memory)
+
+    logger.info(f"✅ Agent 组装完成，已注册工具: {mcp_server.list_tools()}")
+    return client
+
+
+# ── 演示场景 ───────────────────────────────────────────────────
+def run_demo(client: AgentClient) -> None:
+    """运行预设演示场景，展示各工具的完整调用链路"""
+    demo_cases = [
+        ("🔢 数学计算", "计算 (100 + 200) × 3 等于多少"),
+        ("🌐 网络搜索", "搜索 Python 最新版本的新特性"),
+        ("📄 文件读取", "读取文件 config.json 的内容"),
+        ("🐍 代码执行", '执行代码 `print("Hello, Agent!")`'),
+    ]
+
+    logger.info("\n" + "═" * 60)
+    logger.info("🎬 开始演示模式，共 4 个场景")
+    logger.info("═" * 60)
+
+    for title, question in demo_cases:
+        logger.info(f"\n{'─' * 55}")
+        logger.info(f"📌 场景: {title}")
+        logger.info(f"{'─' * 55}")
+
+        response = client.chat(question)
+
+        print(f"\n{'─' * 55}")
+        print(f"👤 用户:  {response.user_input}")
+        if response.tool_used:
+            print(f"🔧 工具:  {response.tool_used}")
+            print(f"📤 输出:  {response.tool_output[:120]}...")
+        print(f"🤖 回复:\n{response.final_reply}")
+        print()
+
+    # 打印记忆统计
+    stats = client.get_memory_stats()
+    logger.info(f"\n📊 Memory 统计: {stats}")
+
+
+# ── 交互式对话循环 ─────────────────────────────────────────────
+def run_interactive(client: AgentClient) -> None:
+    """启动交互式命令行对话"""
+    print("\n" + "═" * 60)
+    print("  🤖 Agent Demo — 交互模式")
+    print("  输入 'quit'  → 退出程序")
+    print("  输入 'clear' → 清空会话历史")
+    print("  输入 'stats' → 查看 Memory 统计")
+    print("  输入 'tools' → 查看已注册工具列表")
+    print("═" * 60 + "\n")
+
+    while True:
+        try:
+            user_input = input("👤 你: ").strip()
+        except (KeyboardInterrupt, EOFError):
+            print("\n👋 再见！")
+            break
+
+        if not user_input:
+            continue
+
+        # ── 内置命令 ──────────────────────────────────────────
+        match user_input.lower():
+            case "quit" | "exit":
+                print("👋 再见！")
+                break
+            case "clear":
+                client.clear_session()
+                print("✅ 会话已清空\n")
+                continue
+            case "stats":
+                print(f"📊 Memory 统计: {client.get_memory_stats()}\n")
+                continue
+            case "tools":
+                tools = client.mcp_server.list_tools()
+                print(f"🔧 已注册工具 ({len(tools)} 个): {', '.join(tools)}\n")
+                continue
+
+        # ── 执行 Agent 完整流程 ───────────────────────────────
+        response = client.chat(user_input)
+
+        print(f"\n{'─' * 55}")
+        if response.tool_used:
+            print(f"  🔧 调用工具: {response.tool_used}")
+        print(f"🤖 Agent:\n{response.final_reply}")
+        print(f"{'─' * 55}\n")
+
+
+# ── 主函数 ─────────────────────────────────────────────────────
+def main() -> None:
+    """
+    主函数入口，支持两种运行模式:
+
+        python main.py          →  交互模式（默认）
+        python main.py demo     →  演示模式（自动执行预设场景）
+    """
+    client = build_agent()
+
+    mode = sys.argv[1] if len(sys.argv) > 1 else "interactive"
+
+    if mode == "demo":
+        run_demo(client)
+    else:
+        run_interactive(client)
+
+
+if __name__ == "__main__":
+    main()
\ No newline at end of file
diff --git a/mcp/__init__.py b/mcp/__init__.py
new file mode 100644
index 0000000..912a555
--- /dev/null
+++ b/mcp/__init__.py
@@ -0,0 +1 @@
+"""__init__.py"""
diff --git a/mcp/__pycache__/__init__.cpython-310.pyc b/mcp/__pycache__/__init__.cpython-310.pyc
new file mode 100644
index 0000000..df2f6cc
Binary files /dev/null and b/mcp/__pycache__/__init__.cpython-310.pyc differ
diff --git a/mcp/__pycache__/mcp_protocol.cpython-310.pyc b/mcp/__pycache__/mcp_protocol.cpython-310.pyc
new file mode 100644
index 0000000..1cb948f
Binary files /dev/null and b/mcp/__pycache__/mcp_protocol.cpython-310.pyc differ
diff --git a/mcp/__pycache__/mcp_server.cpython-310.pyc b/mcp/__pycache__/mcp_server.cpython-310.pyc
new file mode 100644
index 0000000..4f95d1e
Binary files /dev/null and b/mcp/__pycache__/mcp_server.cpython-310.pyc differ
diff --git a/mcp/mcp_protocol.py b/mcp/mcp_protocol.py
new file mode 100644
index 0000000..2126da5
--- /dev/null
+++ b/mcp/mcp_protocol.py
@@ -0,0 +1,107 @@
+"""MCP 协议：JSON-RPC 消息定义"""
+"""
+mcp/mcp_protocol.py
+MCP (Model Context Protocol) 协议数据结构定义
+基于 JSON-RPC 2.0 规范封装请求/响应消息体
+"""
+
+import uuid
+from dataclasses import dataclass, field
+from typing import Any
+
+
+# ── MCP 方法常量 ───────────────────────────────────────────────
+class MCPMethod:
+    TOOLS_LIST   = "tools/list"    # 列出所有可用工具
+    TOOLS_CALL   = "tools/call"    # 调用指定工具
+    RESOURCES_READ = "resources/read"  # 读取资源
+
+
+# ── 请求消息 ───────────────────────────────────────────────────
+@dataclass
+class MCPRequest:
+    """
+    MCP 工具调用请求（JSON-RPC 2.0 格式）
+
+    示例:
+        {
+            "jsonrpc": "2.0",
+            "id": "abc-123",
+            "method": "tools/call",
+            "params": {
+                "name": "calculator",
+                "arguments": {"expression": "1+1"}
+            }
+        }
+    """
+    method:  str
+    params:  dict[str, Any]         = field(default_factory=dict)
+    jsonrpc: str                    = "2.0"
+    id:      str                    = field(default_factory=lambda: str(uuid.uuid4())[:8])
+
+    def to_dict(self) -> dict:
+        return {
+            "jsonrpc": self.jsonrpc,
+            "id":      self.id,
+            "method":  self.method,
+            "params":  self.params,
+        }
+
+
+# ── 响应消息 ───────────────────────────────────────────────────
+@dataclass
+class MCPResponse:
+    """
+    MCP 工具调用响应（JSON-RPC 2.0 格式）
+
+    成功示例:
+        {"jsonrpc": "2.0", "id": "abc-123", "result": {"content": [...]}}
+
+    失败示例:
+        {"jsonrpc": "2.0", "id": "abc-123", "error": {"code": -32601, "message": "..."}}
+    """
+    id:      str
+    result:  dict[str, Any] | None  = None
+    error:   dict[str, Any] | None  = None
+    jsonrpc: str                    = "2.0"
+
+    @property
+    def success(self) -> bool:
+        return self.error is None
+
+    @property
+    def content(self) -> str:
+        """提取响应中的文本内容"""
+        if not self.success or not self.result:
+            return self.error.get("message", "Unknown error") if self.error else ""
+        items = self.result.get("content", [])
+        return "\n".join(item.get("text", "") for item in items if item.get("type") == "text")
+
+    def to_dict(self) -> dict:
+        base = {"jsonrpc": self.jsonrpc, "id": self.id}
+        if self.success:
+            base["result"] = self.result
+        else:
+            base["error"] = self.error
+        return base
+
+
+# ── 工具描述 ───────────────────────────────────────────────────
+@dataclass
+class ToolSchema:
+    """
+    工具的元数据描述，用于 LLM 识别和选择工具
+    """
+    name:        str
+    description: str
+    parameters:  dict[str, Any]   # JSON Schema 格式的参数定义
+
+    def to_dict(self) -> dict:
+        return {
+            "name":        self.name,
+            "description": self.description,
+            "inputSchema": {
+                "type":       "object",
+                "properties": self.parameters,
+            },
+        }
\ No newline at end of file
diff --git a/mcp/mcp_server.py b/mcp/mcp_server.py
new file mode 100644
index 0000000..e0734f3
--- /dev/null
+++ b/mcp/mcp_server.py
@@ -0,0 +1,143 @@
+"""MCP 服务器：工具注册 & 调度"""
+"""
+mcp/mcp_server.py
+MCP Server：工具注册中心与调度引擎
+负责管理所有工具的生命周期，处理 JSON-RPC 格式的工具调用请求
+"""
+
+import json
+from typing import Type
+
+from mcp.mcp_protocol import MCPMethod, MCPRequest, MCPResponse, ToolSchema
+from tools.base_tool import BaseTool, ToolResult
+from utils.logger import get_logger
+
+
+class MCPServer:
+    """
+    MCP 服务器核心类
+
+    职责:
+        1. 工具注册（register_tool）
+        2. 工具列表查询（tools/list）
+        3. 工具调用分发（tools/call）
+        4. JSON-RPC 协议封装/解析
+
+    使用示例:
+        server = MCPServer()
+        server.register_tool(CalculatorTool)
+        response = server.handle_request(request)
+    """
+
+    def __init__(self, server_name: str = "AgentMCPServer"):
+        self.server_name = server_name
+        self.logger      = get_logger("MCP")
+        self._registry:  dict[str, BaseTool] = {}   # 工具名 → 工具实例
+
+        self.logger.info(f"🚀 MCP Server [{server_name}] 启动")
+
+    # ── 工具注册 ────────────────────────────────────────────────
+
+    def register_tool(self, tool_class: Type[BaseTool]) -> None:
+        """
+        注册一个工具类到服务器
+
+        Args:
+            tool_class: 继承自 BaseTool 的工具类（传入类本身，不是实例）
+        """
+        instance = tool_class()
+        if not instance.name:
+            raise ValueError(f"工具类 {tool_class.__name__} 未设置 name 属性")
+
+        self._registry[instance.name] = instance
+        self.logger.info(f"📌 注册工具: [{instance.name}] — {instance.description}")
+
+    def register_tools(self, *tool_classes: Type[BaseTool]) -> None:
+        """批量注册多个工具类"""
+        for cls in tool_classes:
+            self.register_tool(cls)
+
+    # ── 请求处理入口 ────────────────────────────────────────────
+
+    def handle_request(self, request: MCPRequest) -> MCPResponse:
+        """
+        处理 MCP 请求的统一入口，根据 method 分发到对应处理器
+
+        Args:
+            request: MCPRequest 实例
+
+        Returns:
+            MCPResponse 实例
+        """
+        self.logger.info(f"📨 收到请求 id={request.id} method={request.method}")
+
+        handlers = {
+            MCPMethod.TOOLS_LIST: self._handle_tools_list,
+            MCPMethod.TOOLS_CALL: self._handle_tools_call,
+        }
+
+        handler = handlers.get(request.method)
+        if handler is None:
+            return self._error_response(request.id, -32601, f"未知方法: {request.method}")
+
+        return handler(request)
+
+    # ── 私有处理器 ──────────────────────────────────────────────
+
+    def _handle_tools_list(self, request: MCPRequest) -> MCPResponse:
+        """处理 tools/list 请求，返回所有已注册工具的 Schema"""
+        schemas = [tool.get_schema().to_dict() for tool in self._registry.values()]
+        self.logger.info(f"📋 返回工具列表，共 {len(schemas)} 个工具")
+        return MCPResponse(
+            id=request.id,
+            result={"tools": schemas},
+        )
+
+    def _handle_tools_call(self, request: MCPRequest) -> MCPResponse:
+        """处理 tools/call 请求，调用指定工具并返回结果"""
+        tool_name = request.params.get("name")
+        arguments = request.params.get("arguments", {})
+
+        # 检查工具是否存在
+        tool = self._registry.get(tool_name)
+        if tool is None:
+            available = list(self._registry.keys())
+            return self._error_response(
+                request.id, -32602,
+                f"工具 [{tool_name}] 不存在，可用工具: {available}"
+            )
+
+        # 执行工具
+        result: ToolResult = tool.safe_execute(**arguments)
+
+        if result.success:
+            return MCPResponse(
+                id=request.id,
+                result={
+                    "content": [{"type": "text", "text": result.output}],
+                    "metadata": result.metadata,
+                },
+            )
+        else:
+            return self._error_response(request.id, -32000, result.output)
+
+    # ── 工具方法 ────────────────────────────────────────────────
+
+    def get_tool_schemas(self) -> list[ToolSchema]:
+        """获取所有工具的 Schema 列表（供 LLM 引擎使用）"""
+        return [tool.get_schema() for tool in self._registry.values()]
+
+    def list_tools(self) -> list[str]:
+        """返回所有已注册工具的名称列表"""
+        return list(self._registry.keys())
+
+    @staticmethod
+    def _error_response(req_id: str, code: int, message: str) -> MCPResponse:
+        """构造标准 JSON-RPC 错误响应"""
+        return MCPResponse(
+            id=req_id,
+            error={"code": code, "message": message},
+        )
+
+    def __repr__(self) -> str:
+        return f"MCPServer(name={self.server_name!r}, tools={self.list_tools()})"
\ No newline at end of file
diff --git a/memory/__init__.py b/memory/__init__.py
new file mode 100644
index 0000000..912a555
--- /dev/null
+++ b/memory/__init__.py
@@ -0,0 +1 @@
+"""__init__.py"""
diff --git a/memory/__pycache__/__init__.cpython-310.pyc b/memory/__pycache__/__init__.cpython-310.pyc
new file mode 100644
index 0000000..099cc76
Binary files /dev/null and b/memory/__pycache__/__init__.cpython-310.pyc differ
diff --git a/memory/__pycache__/memory_store.cpython-310.pyc b/memory/__pycache__/memory_store.cpython-310.pyc
new file mode 100644
index 0000000..79975c8
Binary files /dev/null and b/memory/__pycache__/memory_store.cpython-310.pyc differ
diff --git a/memory/memory_store.py b/memory/memory_store.py
new file mode 100644
index 0000000..9f521d9
--- /dev/null
+++ b/memory/memory_store.py
@@ -0,0 +1,128 @@
+"""记忆模块：对话历史管理"""
+"""
+memory/memory_store.py
+Agent 记忆模块：管理对话历史（短期记忆）与关键信息摘要（长期记忆）
+"""
+
+from collections import deque
+from dataclasses import dataclass, field
+from datetime import datetime
+from typing import Literal
+
+from utils.logger import get_logger
+
+
+# ── 消息数据结构 ───────────────────────────────────────────────
+@dataclass
+class Message:
+    """单条对话消息"""
+    role:      Literal["user", "assistant", "tool"]
+    content:   str
+    timestamp: str = field(default_factory=lambda: datetime.now().strftime("%H:%M:%S"))
+    metadata:  dict = field(default_factory=dict)
+
+    def to_dict(self) -> dict:
+        return {
+            "role":      self.role,
+            "content":   self.content,
+            "timestamp": self.timestamp,
+        }
+
+
+# ── 记忆存储 ───────────────────────────────────────────────────
+class MemoryStore:
+    """
+    对话记忆存储
+
+    短期记忆: 使用 deque 保存最近 N 轮对话，自动滚动淘汰旧消息
+    长期记忆: 保存关键事实摘要（生产环境可替换为向量数据库）
+
+    使用示例:
+        memory = MemoryStore(max_history=10)
+        memory.add_user_message("你好")
+        memory.add_assistant_message("你好！有什么可以帮你？")
+        history = memory.get_history()
+    """
+
+    def __init__(self, max_history: int = 20):
+        """
+        Args:
+            max_history: 短期记忆保留的最大消息条数
+        """
+        self.logger       = get_logger("MEMORY")
+        self.max_history  = max_history
+        self._history:    deque[Message] = deque(maxlen=max_history)
+        self._facts:      list[str]      = []   # 长期记忆：关键事实
+
+        self.logger.info(f"💾 Memory 初始化，最大历史: {max_history} 条")
+
+    # ── 写入接口 ────────────────────────────────────────────────
+
+    def add_user_message(self, content: str) -> None:
+        """记录用户消息"""
+        self._add(Message(role="user", content=content))
+
+    def add_assistant_message(self, content: str) -> None:
+        """记录 Agent 回复"""
+        self._add(Message(role="assistant", content=content))
+
+    def add_tool_result(self, tool_name: str, result: str) -> None:
+        """记录工具调用结果"""
+        self._add(Message(
+            role="tool",
+            content=result,
+            metadata={"tool": tool_name},
+        ))
+
+    def add_fact(self, fact: str) -> None:
+        """向长期记忆中添加关键事实"""
+        self._facts.append(fact)
+        self.logger.debug(f"📌 长期记忆新增: {fact}")
+
+    # ── 读取接口 ────────────────────────────────────────────────
+
+    def get_history(self, last_n: int | None = None) -> list[dict]:
+        """
+        获取对话历史（LLM 上下文格式）
+
+        Args:
+            last_n: 仅返回最近 N 条，None 表示全部
+
+        Returns:
+            消息字典列表，格式: [{"role": ..., "content": ...}, ...]
+        """
+        messages = list(self._history)
+        if last_n:
+            messages = messages[-last_n:]
+        return [m.to_dict() for m in messages]
+
+    def get_facts(self) -> list[str]:
+        """获取所有长期记忆事实"""
+        return list(self._facts)
+
+    def get_context_summary(self) -> str:
+        """生成上下文摘要字符串，供 LLM Prompt 使用"""
+        history = self.get_history(last_n=6)
+        lines   = [f"[{m['role'].upper()}] {m['content'][:80]}" for m in history]
+        return "\n".join(lines) if lines else "（暂无对话历史）"
+
+    # ── 管理接口 ────────────────────────────────────────────────
+
+    def clear_history(self) -> None:
+        """清空短期对话历史"""
+        self._history.clear()
+        self.logger.info("🗑 对话历史已清空")
+
+    def stats(self) -> dict:
+        """返回记忆统计信息"""
+        return {
+            "history_count": len(self._history),
+            "facts_count":   len(self._facts),
+            "max_history":   self.max_history,
+        }
+
+    # ── 私有方法 ────────────────────────────────────────────────
+
+    def _add(self, message: Message) -> None:
+        self._history.append(message)
+        self.logger.debug(f"💬 [{message.role.upper()}] {message.content[:60]}...")
\ No newline at end of file
diff --git a/tools/__init__.py b/tools/__init__.py
new file mode 100644
index 0000000..912a555
--- /dev/null
+++ b/tools/__init__.py
@@ -0,0 +1 @@
+"""__init__.py"""
diff --git a/tools/__pycache__/__init__.cpython-310.pyc b/tools/__pycache__/__init__.cpython-310.pyc
new file mode 100644
index 0000000..f8aa8fe
Binary files /dev/null and b/tools/__pycache__/__init__.cpython-310.pyc differ
diff --git a/tools/__pycache__/base_tool.cpython-310.pyc b/tools/__pycache__/base_tool.cpython-310.pyc
new file mode 100644
index 0000000..4b303e2
Binary files /dev/null and b/tools/__pycache__/base_tool.cpython-310.pyc differ
diff --git a/tools/__pycache__/calculator.cpython-310.pyc b/tools/__pycache__/calculator.cpython-310.pyc
new file mode 100644
index 0000000..d7b5b4f
Binary files /dev/null and b/tools/__pycache__/calculator.cpython-310.pyc differ
diff --git a/tools/__pycache__/code_executor.cpython-310.pyc b/tools/__pycache__/code_executor.cpython-310.pyc
new file mode 100644
index 0000000..c69976c
Binary files /dev/null and b/tools/__pycache__/code_executor.cpython-310.pyc differ
diff --git a/tools/__pycache__/file_reader.cpython-310.pyc b/tools/__pycache__/file_reader.cpython-310.pyc
new file mode 100644
index 0000000..2f8fd37
Binary files /dev/null and b/tools/__pycache__/file_reader.cpython-310.pyc differ
diff --git a/tools/__pycache__/web_search.cpython-310.pyc b/tools/__pycache__/web_search.cpython-310.pyc
new file mode 100644
index 0000000..81f3f95
Binary files /dev/null and b/tools/__pycache__/web_search.cpython-310.pyc differ
diff --git a/tools/base_tool.py b/tools/base_tool.py
new file mode 100644
index 0000000..ef4f48a
--- /dev/null
+++ b/tools/base_tool.py
@@ -0,0 +1,82 @@
+"""
+tools/base_tool.py
+所有工具的抽象基类，定义统一接口规范
+"""
+
+from abc import ABC, abstractmethod
+from dataclasses import dataclass
+from typing import Any
+
+from mcp.mcp_protocol import ToolSchema
+from utils.logger import get_logger
+
+
+@dataclass
+class ToolResult:
+    """工具执行结果的统一封装"""
+    success:  bool
+    output:   str
+    metadata: dict[str, Any] = None
+
+    def __post_init__(self):
+        self.metadata = self.metadata or {}
+
+
+class BaseTool(ABC):
+    """
+    工具基类：所有具体工具必须继承此类并实现 execute() 方法
+
+    子类示例:
+        class MyTool(BaseTool):
+            name        = "my_tool"
+            description = "这是我的工具"
+            parameters  = {"input": {"type": "string", "description": "输入内容"}}
+
+            def execute(self, **kwargs) -> ToolResult:
+                return ToolResult(success=True, output=f"处理结果: {kwargs['input']}")
+    """
+
+    # 子类必须定义的类属性
+    name:        str = ""
+    description: str = ""
+    parameters:  dict[str, Any] = {}
+
+    def __init__(self):
+        self.logger = get_logger("TOOL")
+
+    @abstractmethod
+    def execute(self, **kwargs) -> ToolResult:
+        """
+        执行工具逻辑（子类必须实现）
+
+        Args:
+            **kwargs: 工具参数，与 parameters 中定义的字段对应
+
+        Returns:
+            ToolResult 实例
+        """
+        ...
+
+    def get_schema(self) -> ToolSchema:
+        """返回工具的 MCP Schema 描述"""
+        return ToolSchema(
+            name=self.name,
+            description=self.description,
+            parameters=self.parameters,
+        )
+
+    def safe_execute(self, **kwargs) -> ToolResult:
+        """
+        带异常捕获的安全执行入口，由 MCP Server 调用
+
+        Returns:
+            ToolResult，失败时 success=False 并携带错误信息
+        """
+        self.logger.info(f"▶ 执行工具 [{self.name}]，参数: {kwargs}")
+        try:
+            result = self.execute(**kwargs)
+            self.logger.info(f"✅ 工具 [{self.name}] 执行成功")
+            return result
+        except Exception as exc:
+            self.logger.error(f"❌ 工具 [{self.name}] 执行失败: {exc}")
+            return ToolResult(success=False, output=f"工具执行异常: {exc}")
\ No newline at end of file
diff --git a/tools/calculator.py b/tools/calculator.py
new file mode 100644
index 0000000..76e5d4f
--- /dev/null
+++ b/tools/calculator.py
@@ -0,0 +1,64 @@
+"""计算器工具"""
+# ════════════════════════════════════════════════════════════════
+# tools/calculator.py   —   数学计算工具
+# ════════════════════════════════════════════════════════════════
+"""
+tools/calculator.py
+安全的数学表达式计算工具（使用 ast 模块避免 eval 注入风险）
+"""
+
+import ast
+import operator
+from tools.base_tool import BaseTool, ToolResult
+
+
+class CalculatorTool(BaseTool):
+    name        = "calculator"
+    description = "计算数学表达式，支持加减乘除、幂运算、括号等"
+    parameters  = {
+        "expression": {
+            "type":        "string",
+            "description": "数学表达式，例如 '(1+2)*3' 或 '2**10'",
+        }
+    }
+
+    # 允许的运算符白名单（防止注入）
+    _OPERATORS = {
+        ast.Add:  operator.add,
+        ast.Sub:  operator.sub,
+        ast.Mult: operator.mul,
+        ast.Div:  operator.truediv,
+        ast.Pow:  operator.pow,
+        ast.Mod:  operator.mod,
+        ast.USub: operator.neg,
+    }
+
+    def execute(self, expression: str, **_) -> ToolResult:
+        try:
+            tree   = ast.parse(expression, mode="eval")
+            result = self._eval_node(tree.body)
+            return ToolResult(
+                success=True,
+                output=f"{expression} = {result}",
+                metadata={"expression": expression, "result": result},
+            )
+        except (ValueError, TypeError, ZeroDivisionError) as exc:
+            return ToolResult(success=False, output=f"计算错误: {exc}")
+
+    def _eval_node(self, node: ast.AST) -> float:
+        """递归解析 AST 节点"""
+        match node:
+            case ast.Constant(value=v) if isinstance(v, (int, float)):
+                return v
+            case ast.BinOp(left=left, op=op, right=right):
+                fn = self._OPERATORS.get(type(op))
+                if fn is None:
+                    raise ValueError(f"不支持的运算符: {type(op).__name__}")
+                return fn(self._eval_node(left), self._eval_node(right))
+            case ast.UnaryOp(op=op, operand=operand):
+                fn = self._OPERATORS.get(type(op))
+                if fn is None:
+                    raise ValueError(f"不支持的一元运算符: {type(op).__name__}")
+                return fn(self._eval_node(operand))
+            case _:
+                raise ValueError(f"不支持的表达式节点: {type(node).__name__}")
\ No newline at end of file
diff --git a/tools/code_executor.py b/tools/code_executor.py
new file mode 100644
index 0000000..11be540
--- /dev/null
+++ b/tools/code_executor.py
@@ -0,0 +1,60 @@
+"""代码执行工具"""
+
+# ════════════════════════════════════════════════════════════════
+# tools/code_executor.py   —   代码执行工具
+# ════════════════════════════════════════════════════════════════
+"""
+tools/code_executor.py
+沙箱代码执行工具：在受限环境中运行 Python 代码片段
+"""
+
+import io
+import contextlib
+import time
+from tools.base_tool import BaseTool, ToolResult
+
+
+class CodeExecutorTool(BaseTool):
+    name        = "code_executor"
+    description = "在沙箱环境中执行 Python 代码片段，返回标准输出"
+    parameters  = {
+        "code": {
+            "type":        "string",
+            "description": "要执行的 Python 代码",
+        },
+        "timeout": {
+            "type":        "integer",
+            "description": "超时时间（秒），默认 5",
+        },
+    }
+
+    # 沙箱：仅允许安全的内置函数
+    _SAFE_BUILTINS = {
+        "print": print, "range": range, "len": len,
+        "int": int, "float": float, "str": str, "list": list,
+        "dict": dict, "tuple": tuple, "set": set, "bool": bool,
+        "abs": abs, "max": max, "min": min, "sum": sum,
+        "enumerate": enumerate, "zip": zip, "map": map,
+        "sorted": sorted, "reversed": reversed,
+    }
+
+    def execute(self, code: str, timeout: int = 5, **_) -> ToolResult:
+        stdout_buf = io.StringIO()
+        start_time = time.perf_counter()
+
+        try:
+            # 重定向 stdout，捕获 print 输出
+            with contextlib.redirect_stdout(stdout_buf):
+                exec(  # noqa: S102
+                    compile(code, "<agent_sandbox>", "exec"),
+                    {"__builtins__": self._SAFE_BUILTINS},
+                )
+            elapsed = (time.perf_counter() - start_time) * 1000
+            output  = stdout_buf.getvalue() or "(无输出)"
+            return ToolResult(
+                success=True,
+                output=f"执行成功 ({elapsed:.1f}ms):\n{output}",
+                metadata={"elapsed_ms": elapsed},
+            )
+        except Exception as exc:
+            return ToolResult(success=False, output=f"执行错误: {type(exc).__name__}: {exc}")
\ No newline at end of file
diff --git a/tools/file_reader.py b/tools/file_reader.py
new file mode 100644
index 0000000..7942316
--- /dev/null
+++ b/tools/file_reader.py
@@ -0,0 +1,65 @@
+"""文件读取工具"""
+
+
+# ════════════════════════════════════════════════════════════════
+# tools/file_reader.py   —   文件读取工具
+# ════════════════════════════════════════════════════════════════
+"""
+tools/file_reader.py
+本地文件读取工具，支持文本文件，限制读取路径防止越权
+"""
+
+from pathlib import Path
+from tools.base_tool import BaseTool, ToolResult
+
+
+# 允许读取的根目录（沙箱限制）
+_ALLOWED_ROOT = Path("./workspace")
+
+
+class FileReaderTool(BaseTool):
+    name        = "file_reader"
+    description = "读取本地文件内容，仅限 workspace/ 目录下的文件"
+    parameters  = {
+        "path": {
+            "type":        "string",
+            "description": "文件路径，相对于 workspace/ 目录",
+        },
+        "encoding": {
+            "type":        "string",
+            "description": "文件编码，默认 utf-8",
+        },
+    }
+
+    def execute(self, path: str, encoding: str = "utf-8", **_) -> ToolResult:
+        _ALLOWED_ROOT.mkdir(exist_ok=True)
+
+        # 路径安全检查：防止目录穿越攻击
+        target = (_ALLOWED_ROOT / path).resolve()
+        if not str(target).startswith(str(_ALLOWED_ROOT.resolve())):
+            return ToolResult(success=False, output=f"❌ 拒绝访问: 路径超出允许范围")
+
+        if not target.exists():
+            # Demo 模式：自动创建示例文件
+            self._create_demo_file(target)
+
+        try:
+            content = target.read_text(encoding=encoding)
+            return ToolResult(
+                success=True,
+                output=f"文件 [{path}] 内容:\n{content}",
+                metadata={"path": str(target), "size": target.stat().st_size},
+            )
+        except OSError as exc:
+            return ToolResult(success=False, output=f"读取失败: {exc}")
+
+    @staticmethod
+    def _create_demo_file(path: Path) -> None:
+        """自动创建演示用文件"""
+        path.parent.mkdir(parents=True, exist_ok=True)
+        path.write_text(
+            '{\n  "app": "AgentDemo",\n  "version": "1.0.0",\n'
+            '  "llm": "claude-sonnet-4-6",\n  "tools": ["calculator", "web_search"]\n}\n',
+            encoding="utf-8",
+        )
+
diff --git a/tools/web_search.py b/tools/web_search.py
new file mode 100644
index 0000000..3b2a318
--- /dev/null
+++ b/tools/web_search.py
@@ -0,0 +1,65 @@
+"""网络搜索工具"""
+
+# ════════════════════════════════════════════════════════════════
+# tools/web_search.py   —   网络搜索工具（模拟）
+# ════════════════════════════════════════════════════════════════
+"""
+tools/web_search.py
+网络搜索工具（Demo 中使用模拟数据，生产环境可替换为真实 API）
+"""
+
+import time
+from tools.base_tool import BaseTool, ToolResult
+
+
+# 模拟搜索结果数据库
+_MOCK_RESULTS: dict[str, list[dict]] = {
+    "天气": [
+        {"title": "今日天气预报",    "snippet": "晴转多云，气温 15°C ~ 24°C，东南风 3 级"},
+        {"title": "未来 7 天天气",   "snippet": "本周整体晴好，周末有小雨"},
+    ],
+    "python": [
+        {"title": "Python 官方文档", "snippet": "Python 3.12 新特性：改进的错误提示、更快的启动速度"},
+        {"title": "Python 教程",     "snippet": "从零开始学 Python，包含 300+ 实战案例"},
+    ],
+}
+_DEFAULT_RESULTS = [
+    {"title": "搜索结果 1", "snippet": "找到相关内容，请查看详情"},
+    {"title": "搜索结果 2", "snippet": "更多相关信息可通过链接访问"},
+]
+
+
+class WebSearchTool(BaseTool):
+    name        = "web_search"
+    description = "在互联网上搜索信息，返回相关网页摘要"
+    parameters  = {
+        "query": {
+            "type":        "string",
+            "description": "搜索关键词或问题",
+        },
+        "max_results": {
+            "type":        "integer",
+            "description": "返回结果数量，默认 3",
+        },
+    }
+
+    def execute(self, query: str, max_results: int = 3, **_) -> ToolResult:
+        time.sleep(0.1)  # 模拟网络延迟
+
+        # 关键词匹配模拟结果
+        results = _DEFAULT_RESULTS
+        for keyword, data in _MOCK_RESULTS.items():
+            if keyword in query:
+                results = data
+                break
+
+        results = results[:max_results]
+        formatted = "\n".join(
+            f"[{i+1}] {r['title']}\n    {r['snippet']}"
+            for i, r in enumerate(results)
+        )
+        return ToolResult(
+            success=True,
+            output=f"搜索「{query}」，共 {len(results)} 条结果:\n{formatted}",
+            metadata={"query": query, "count": len(results)},
+        )
diff --git a/utils/__pycache__/logger.cpython-310.pyc b/utils/__pycache__/logger.cpython-310.pyc
new file mode 100644
index 0000000..bd83d77
Binary files /dev/null and b/utils/__pycache__/logger.cpython-310.pyc differ
diff --git a/utils/logger.py b/utils/logger.py
new file mode 100644
index 0000000..90191a6
--- /dev/null
+++ b/utils/logger.py
@@ -0,0 +1,96 @@
+"""日志工具"""
+"""
+utils/logger.py
+统一日志模块，支持彩色终端输出与文件记录
+"""
+
+import logging
+import sys
+from datetime import datetime
+from pathlib import Path
+
+
+# ── ANSI 颜色常量 ──────────────────────────────────────────────
+class Color:
+    RESET   = "\033[0m"
+    BOLD    = "\033[1m"
+    CYAN    = "\033[96m"
+    GREEN   = "\033[92m"
+    YELLOW  = "\033[93m"
+    RED     = "\033[91m"
+    MAGENTA = "\033[95m"
+    BLUE    = "\033[94m"
+    GREY    = "\033[90m"
+
+
+# ── 自定义彩色 Formatter ───────────────────────────────────────
+class ColorFormatter(logging.Formatter):
+    LEVEL_COLORS = {
+        logging.DEBUG:    Color.GREY,
+        logging.INFO:     Color.CYAN,
+        logging.WARNING:  Color.YELLOW,
+        logging.ERROR:    Color.RED,
+        logging.CRITICAL: Color.MAGENTA,
+    }
+
+    COMPONENT_COLORS = {
+        "CLIENT":  Color.BLUE,
+        "LLM":     Color.GREEN,
+        "MCP":     Color.YELLOW,
+        "TOOL":    Color.MAGENTA,
+        "MEMORY":  Color.CYAN,
+        "SYSTEM":  Color.GREY,
+    }
+
+    def format(self, record: logging.LogRecord) -> str:
+        level_color = self.LEVEL_COLORS.get(record.levelno, Color.RESET)
+        time_str    = datetime.now().strftime("%H:%M:%S.%f")[:-3]
+
+        # 从 logger 名称提取组件标签，例如 "agent.CLIENT"
+        component   = record.name.split(".")[-1].upper()
+        comp_color  = self.COMPONENT_COLORS.get(component, Color.RESET)
+
+        prefix = (
+            f"{Color.GREY}[{time_str}]{Color.RESET} "
+            f"{comp_color}{Color.BOLD}[{component:6s}]{Color.RESET} "
+            f"{level_color}{record.getMessage()}{Color.RESET}"
+        )
+        return prefix
+
+
+# ── Logger 工厂函数 ────────────────────────────────────────────
+def get_logger(component: str, level: int = logging.DEBUG) -> logging.Logger:
+    """
+    获取指定组件的 Logger 实例。
+
+    Args:
+        component: 组件名称，如 "CLIENT"、"LLM"、"MCP"
+        level:     日志级别，默认 DEBUG
+
+    Returns:
+        配置好的 Logger 实例
+    """
+    logger = logging.getLogger(f"agent.{component}")
+    logger.setLevel(level)
+
+    # 避免重复添加 Handler
+    if logger.handlers:
+        return logger
+
+    # 终端 Handler（彩色）
+    console_handler = logging.StreamHandler(sys.stdout)
+    console_handler.setFormatter(ColorFormatter())
+    logger.addHandler(console_handler)
+
+    # 文件 Handler（纯文本）
+    log_dir = Path("logs")
+    log_dir.mkdir(exist_ok=True)
+    file_handler = logging.FileHandler(log_dir / "agent.log", encoding="utf-8")
+    file_handler.setFormatter(
+        logging.Formatter("[%(asctime)s] [%(name)s] %(levelname)s: %(message)s")
+    )
+    logger.addHandler(file_handler)
+
+    # 防止日志向上传播到 root logger
+    logger.propagate = False
+    return logger
\ No newline at end of file