![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/05\/20250521233123-682e624b5e83c.png\")
<\/p>\n\u5728\u4eba\u5de5\u667a\u80fd\u5e94\u7528\u65e5\u76ca\u666e\u53ca\u7684\u4eca\u5929，\u9690\u79c1\u4fdd\u62a4\u4e0e\u6570\u636e\u4e3b\u6743\u6210\u4e3a\u91cd\u8981\u6311\u6218\u3002\u4f20\u7edf\u7684AI\u6a21\u578b\u4f9d\u8d56\u5916\u90e8\u670d\u52a1，\u5bfc\u81f4\u79c1\u6709\u77e5\u8bc6\u9762\u4e34\u6cc4\u9732\u98ce\u9669\u3002\u672c\u6587\u5c06\u8be6\u7ec6\u4ecb\u7ecd\u5982\u4f55\u6784\u5efa\u4e00\u4e2a\u5b8c\u5168\u672c\u5730\u5316\u7684AI\u667a\u80fd\u4f53，\u901a\u8fc7\u81ea\u5b9a\u4e49\u7684Model Context Protocol（MCP）\u670d\u52a1\u5668\u5b9e\u73b0\u77e5\u8bc6\u9694\u79bb，\u5e76\u7ed3\u5408Llama 3.2\u8f7b\u91cf\u7ea7\u6a21\u578b\u5b9e\u73b0\u5de5\u5177\u8c03\u7528\u80fd\u529b\u3002\u8fd9\u4e00\u65b9\u6848\u4e0d\u4ec5\u786e\u4fdd\u6570\u636e\u5b8c\u5168\u5728\u672c\u5730\u8fd0\u884c，\u8fd8\u80fd\u901a\u8fc7\u5de5\u5177\u8c03\u7528\u4e0e\u79c1\u6709\u77e5\u8bc6\u5e93\u6df1\u5ea6\u4ea4\u4e92，\u4e3a\u672c\u5730\u5316\u667a\u80fd\u5e94\u7528\u63d0\u4f9b\u4e86\u53ef\u884c\u8def\u5f84\u3002<\/p>\n<\/p>\n
<\/p>\n
\u5728\u4e4b\u524d\u7684\u6587\u7ae0\u4e2d，\u7b14\u8005\u6784\u5efa\u4e86\u4e00\u4e2a\u81ea\u5b9a\u4e49MCP\u670d\u52a1\u5668，\u5176\u6838\u5fc3\u76ee\u6807\u662f\u5b9e\u73b0\u4e09\u5927\u529f\u80fd：<\/p>\n
\u907f\u514dAI\u6a21\u578b\u5bf9\u6587\u4ef6\u7cfb\u7edf\u8fdb\u884c\u5199\u5165\u64cd\u4f5c，\u9632\u6b62\u6570\u636e\u7be1\u6539\u3002<\/p>\n<\/li>\n
\u9690\u85cf\u6587\u4ef6\u76ee\u5f55\u7ed3\u6784，\u4ec5\u5411\u6a21\u578b\u66b4\u9732\u5fc5\u8981\u7684\u77e5\u8bc6\u5185\u5bb9\u3002<\/p>\n<\/li>\n
\u901a\u8fc7\u81ea\u4e3b\u5b9e\u73b0MCP\u534f\u8bae，\u6df1\u5165\u638c\u63e1\u5176\u5de5\u4f5c\u539f\u7406\u3002<\/p>\n<\/li>\n
\u8be5\u670d\u52a1\u5668\u901a\u8fc7\u6807\u51c6\u8f93\u5165\u8f93\u51fa（stdio）\u4e0e\u5916\u90e8\u901a\u4fe1，\u80fd\u591f\u8fde\u63a5Obsidian\u77e5\u8bc6\u5e93，\u63d0\u4f9b\u5de5\u5177\u8c03\u7528\u63a5\u53e3\u3002\u4f8b\u5982，list_knowledges\u5de5\u5177\u7528\u4e8e\u83b7\u53d6\u77e5\u8bc6\u5e93\u4e2d\u7684\u6587\u4ef6\u5217\u8868，get_knowledge_by_uri\u5de5\u5177\u901a\u8fc7URI\u68c0\u7d22\u5177\u4f53\u77e5\u8bc6\u5185\u5bb9\u3002\u8fd9\u4e9b\u5de5\u5177\u4e3a\u540e\u7eed\u667a\u80fd\u4f53\u7684\u6784\u5efa\u5960\u5b9a\u4e86\u57fa\u7840\u3002<\/p>\n
\u5c3d\u7ba1\u73b0\u6709MCP\u670d\u52a1\u5668\u5df2\u5b9e\u73b0\u77e5\u8bc6\u9694\u79bb，\u4f46\u4f9d\u8d56\u5916\u90e8AI\u6a21\u578b（\u5982Claude）\u4ecd\u5b58\u5728\u4e24\u5927\u95ee\u9898：<\/p>\n
Claude\u7b49\u670d\u52a1\u9700\u4ed8\u8d39\u8ba2\u9605，\u514d\u8d39\u7248\u672c\u529f\u80fd\u53d7\u9650\u3002<\/p>\n<\/li>\n
\u79c1\u6709\u77e5\u8bc6\u9700\u4f20\u8f93\u81f3\u5916\u90e8\u670d\u52a1\u5668，\u5b58\u5728\u6cc4\u9732\u9690\u60a3\u3002<\/p>\n<\/li>\n<\/ul>\n
\u56e0\u6b64，\u6784\u5efa\u5b8c\u5168\u672c\u5730\u5316\u7684\u667a\u80fd\u4f53\u6210\u4e3a\u5fc5\u7136\u9009\u62e9\u3002\u6838\u5fc3\u76ee\u6807\u5305\u62ec：<\/p>\n
\u5b9e\u73b0MCP\u5ba2\u6237\u7aef，\u4e0e\u81ea\u5b9a\u4e49\u670d\u52a1\u5668\u901a\u4fe1\u3002<\/p>\n<\/li>\n
\u96c6\u6210\u672c\u5730\u8fd0\u884c\u7684LLM\u6a21\u578b，\u66ff\u4ee3\u5916\u90e8\u670d\u52a1\u3002<\/p>\n<\/li>\n
\u6784\u5efa\u667a\u80fd\u4f53\u6846\u67b6，\u7ed3\u5408MCP\u5de5\u5177\u4e0eLLM\u5b9e\u73b0\u95ee\u7b54\u903b\u8f91\u3002<\/p>\n<\/li>\n
\u5728\u667a\u80fd\u4f53\u5f00\u53d1\u4e2d，\u8bed\u8a00\u6a21\u578b\u662f\u6838\u5fc3\u201c\u5927\u8111\u201d\u3002\u8003\u8651\u5230\u672c\u5730\u5316\u8fd0\u884c\u9700\u6c42，\u9700\u9009\u62e9\u8f7b\u91cf\u7ea7\u4e14\u652f\u6301\u5de5\u5177\u8c03\u7528\u7684\u6a21\u578b\u3002Llama 3.2\u7cfb\u5217\u76841B\/3B\u6a21\u578b\u6210\u4e3a\u7406\u60f3\u9009\u62e9，\u5176\u7279\u70b9\u5305\u62ec：<\/p>\n
\u53ef\u5728\u672c\u5730GPU\/CPU\u8fd0\u884c，\u65e0\u9700\u4e91\u7aef\u8d44\u6e90\u3002<\/p>\n<\/li>\n
\u5185\u7f6e\u5bf9\u51fd\u6570\u8c03\u7528\u7684\u7406\u89e3\u80fd\u529b，\u7b26\u5408MCP\u534f\u8bae\u9700\u6c42\u3002<\/p>\n<\/li>\n
\u901a\u8fc7\u7ed3\u6784\u5316\u526a\u679d\u4e0e\u77e5\u8bc6\u84b8\u998f，\u5728\u6a21\u578b\u5927\u5c0f\u4e0e\u63a8\u7406\u80fd\u529b\u95f4\u53d6\u5f97\u5e73\u8861\u3002<\/p>\n<\/li>\n<\/ul>\n
Meta\u5b98\u65b9\u6570\u636e\u663e\u793a，Llama 3.2 3B\u6a21\u578b\u5728\u4fdd\u6301\u8f83\u5c0f\u4f53\u79ef\u7684\u540c\u65f6，\u80fd\u591f\u5904\u7406\u590d\u6742\u6307\u4ee4\u5e76\u751f\u6210\u9ad8\u8d28\u91cf\u54cd\u5e94\u3002\u4f8b\u5982，\u5728\u5de5\u5177\u8c03\u7528\u573a\u666f\u4e2d，\u8be5\u6a21\u578b\u53ef\u89e3\u6790\u51fd\u6570\u53c2\u6570\u5e76\u751f\u6210\u6b63\u786e\u7684\u8c03\u7528\u683c\u5f0f，\u5c3d\u7ba1\u5176\u591a\u8f6e\u5bf9\u8bdd\u80fd\u529b\u7a0d\u900a\u4e8e70B\u4ee5\u4e0a\u7684\u5927\u578b\u6a21\u578b\u3002<\/p>\n
Llama\u6a21\u578b\u7684\u5de5\u5177\u8c03\u7528\u57fa\u4e8e\u7279\u5b9a\u7684\u63d0\u793a\u683c\u5f0f\u4e0e\u89d2\u8272\u673a\u5236，\u6838\u5fc3\u6d41\u7a0b\u5982\u4e0b：<\/p>\n
\u7528\u6237\u95ee\u9898\u89e6\u53d1\u6a21\u578b\u5224\u65ad\u662f\u5426\u9700\u8981\u5de5\u5177\u3002\u4f8b\u5982，\u67e5\u8be2\u7528\u6237ID\u4e3a7890\u7684\u540d\u79f0\u65f6，\u6a21\u578b\u751f\u6210\u5de5\u5177\u8c03\u7528\u8868\u8fbe\u5f0f[get_user_name(user_id=7890)]\u3002<\/p>\n<\/li>\n
\u5e94\u7528\u89e3\u6790\u8c03\u7528\u8868\u8fbe\u5f0f，\u901a\u8fc7MCP\u5ba2\u6237\u7aef\u6267\u884c\u5de5\u5177，\u5c06\u7ed3\u679c（\u5982{"output": "Hyunjong Lee"}）\u4ee5ipython\u89d2\u8272\u8fd4\u56de\u6a21\u578b\u3002<\/p>\n<\/li>\n
\u6a21\u578b\u7ed3\u5408\u5de5\u5177\u8f93\u51fa\u751f\u6210\u6700\u7ec8\u56de\u7b54，\u5982\u201cThe name of user who has the ID is Hyunjong Lee\u201d\u3002<\/p>\n<\/li>\n
\u9700\u8981\u6ce8\u610f\u7684\u662f，\u8f7b\u91cf\u7ea7\u6a21\u578b（\u59823B）\u5728\u5904\u7406\u591a\u8f6e\u5de5\u5177\u8c03\u7528\u65f6\u53ef\u80fd\u4e0d\u7a33\u5b9a\u3002Meta\u5efa\u8bae，\u5bf9\u4e8e\u590d\u6742\u5bf9\u8bdd\u573a\u666f，\u4f18\u5148\u4f7f\u752870B\u4ee5\u4e0a\u6a21\u578b，\u4f46\u5728\u5355\u8f6e\u6216\u7b80\u5355\u591a\u8f6e\u8c03\u7528\u4e2d，3B\u6a21\u578b\u4ecd\u53ef\u80dc\u4efb\u3002<\/p>\n
\u667a\u80fd\u4f53\u7684\u6574\u4f53\u67b6\u6784\u5305\u542b\u4e09\u5927\u6838\u5fc3\u7ec4\u4ef6：MCP\u5ba2\u6237\u7aef\u4e0e\u7ba1\u7406\u5668\u3001LLM\u6a21\u578b\u63a5\u53e3\u3001\u667a\u80fd\u4f53\u903b\u8f91\u5c42\u3002\u4ee5\u4e0b\u662f\u5404\u90e8\u5206\u7684\u8be6\u7ec6\u5b9e\u73b0\u3002<\/p>\n
\u4f7f\u7528Python\u7684MCP SDK\u6784\u5efa\u5ba2\u6237\u7aef，\u901a\u8fc7\u6807\u51c6\u8f93\u5165\u8f93\u51fa\u4e0e\u670d\u52a1\u5668\u8fdb\u7a0b\u901a\u4fe1\u3002\u6838\u5fc3\u7c7bMCPClient\u8d1f\u8d23\u8fde\u63a5\u670d\u52a1\u5668\u3001\u521d\u59cb\u5316\u4f1a\u8bdd\u5e76\u6267\u884c\u5de5\u5177\u8c03\u7528：<\/p>\n
class MCPClient:
\n\u00a0 \u00a0 async def connect_to_server(self, server_script_path):
\n\u00a0 \u00a0 \u00a0 \u00a0 server_params = StdioServerParameters(command="python", args=[server_script_path])
\n\u00a0 \u00a0 \u00a0 \u00a0 stdio_transport = await self.exit_stack.enter_async_context(stdio_client(server_params))
\n\u00a0 \u00a0 \u00a0 \u00a0 self.read, self.write = stdio_transport
\n\u00a0 \u00a0 \u00a0 \u00a0 self.session = await self.exit_stack.enter_async_context(ClientSession(self.read, self.write))
\n\u00a0 \u00a0 \u00a0 \u00a0 init_result = await self.session.initialize()
\n\u00a0 \u00a0 \u00a0 \u00a0 self.name = f"{init_result.serverInfo.name}(v{init_result.serverInfo.version})"<\/p>\n
\u00a0 \u00a0 async def call_tool(self, name, args):
\n\u00a0 \u00a0 \u00a0 \u00a0 response = await self.session.call_tool(name, args)
\n\u00a0 \u00a0 \u00a0 \u00a0 return response.isError, response.content<\/p>\n
\u5ba2\u6237\u7aef\u9075\u5faaMCP\u63e1\u624b\u6d41\u7a0b：\u9996\u5148\u53d1\u9001\u521d\u59cb\u5316\u8bf7\u6c42，\u83b7\u53d6\u670d\u52a1\u5668\u4fe1\u606f（\u5982\u534f\u8bae\u7248\u672c\u3001\u5de5\u5177\u5217\u8868），\u7136\u540e\u901a\u8fc7list_tools\u3001list_resources\u7b49\u65b9\u6cd5\u679a\u4e3e\u53ef\u7528\u8d44\u6e90，\u901a\u8fc7call_tool\u6267\u884c\u5177\u4f53\u5de5\u5177\u3002<\/p>\n
\u4e3a\u652f\u6301\u591a\u670d\u52a1\u5668\u7ba1\u7406，\u8bbe\u8ba1MCPClientManager\u7c7b，\u8d1f\u8d23\u5ba2\u6237\u7aef\u5b9e\u4f8b\u7684\u521b\u5efa\u3001\u9500\u6bc1\u53ca\u5de5\u5177\u6620\u5c04：<\/p>\n
class MCPClientManager:
\n\u00a0 \u00a0 def __init__(self):
\n\u00a0 \u00a0 \u00a0 \u00a0 self.clients = []
\n\u00a0 \u00a0 \u00a0 \u00a0 self.tool_map = {} \u00a0# \u5de5\u5177\u540d\u5230\u5ba2\u6237\u7aef\u7d22\u5f15\u7684\u6620\u5c04<\/p>\n
\u00a0 \u00a0 async def init_mcp_client(self, server_paths):
\n\u00a0 \u00a0 \u00a0 \u00a0 for path in server_paths:
\n\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 client = MCPClient()
\n\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 await client.connect_to_server(path)
\n\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 self.clients.append(client)
\n\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 tools = await client.list_tools()
\n\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 for tool in tools:
\n\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 self.tool_map[tool.name] = len(self.clients) – 1<\/p>\n
\u00a0 \u00a0 async def call_tool(self, name, params):
\n\u00a0 \u00a0 \u00a0 \u00a0 idx = self.tool_map.get(name, -1)
\n\u00a0 \u00a0 \u00a0 \u00a0 if idx == -1:
\n\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 raise Exception(f"Tool {name} not found")
\n\u00a0 \u00a0 \u00a0 \u00a0 return await self.clients[idx].call_tool(name, params)<\/p>\n
\u7ba1\u7406\u5668\u7ef4\u62a4\u5de5\u5177\u4e0e\u5ba2\u6237\u7aef\u7684\u6620\u5c04\u5173\u7cfb，\u786e\u4fdd\u5de5\u5177\u8c03\u7528\u8bf7\u6c42\u6b63\u786e\u8def\u7531\u81f3\u5bf9\u5e94\u7684\u670d\u52a1\u5668\u5b9e\u4f8b\u3002<\/p>\n
\u4f7f\u7528Llama.cpp\u5e93\u5728\u672c\u5730\u8fd0\u884cLlama\u6a21\u578b，\u6b65\u9aa4\u5982\u4e0b：<\/p>\n
Python\u63a5\u53e3\u5c01\u88c5：\u901a\u8fc7\u81ea\u5b9a\u4e49\u7c7bLlamaCPP\u5305\u88c5Llama.cpp\u7684\u63a8\u7406\u63a5\u53e3，\u652f\u6301\u63d0\u793a\u751f\u6210\u4e0e\u54cd\u5e94\u89e3\u6790\u3002<\/p>\n
class LlamaCPP:
\n\u00a0 \u00a0 def __init__(self, model_path):
\n\u00a0 \u00a0 \u00a0 \u00a0 self.model = Llama(model_path=model_path, n_ctx=1024)<\/p>\n
\u00a0 \u00a0 def generate(self, prompt, max_tokens=512):
\n\u00a0 \u00a0 \u00a0 \u00a0 output = self.model(prompt, max_tokens=max_tokens)
\n\u00a0 \u00a0 \u00a0 \u00a0 return output["choices"][0]["text"].strip()\n <\/li>\n<\/ul>\n
\u4e3a\u9002\u914dLlama\u7684\u63d0\u793a\u683c\u5f0f，\u8bbe\u8ba1LlamaMessage\u4e0eLlamaPrompt\u7c7b，\u8d1f\u8d23\u6d88\u606f\u683c\u5f0f\u5316\u4e0e\u5bf9\u8bdd\u5386\u53f2\u7ef4\u62a4：<\/p>\n
class LlamaMessage:
\n\u00a0 \u00a0 def __init__(self, role, content="", tool_scheme=""):
\n\u00a0 \u00a0 \u00a0 \u00a0 self.role = role
\n\u00a0 \u00a0 \u00a0 \u00a0 self.content = content
\n\u00a0 \u00a0 \u00a0 \u00a0 self.tool_scheme = tool_scheme<\/p>\n
\u00a0 \u00a0 def template(self, tool_enabled=False):
\n\u00a0 \u00a0 \u00a0 \u00a0 prompt = f"<|start_header_id|>{self.role}<|end_header_id|>"
\n\u00a0 \u00a0 \u00a0 \u00a0 if tool_enabled and self.tool_scheme:
\n\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 prompt += self.tool_scheme
\n\u00a0 \u00a0 \u00a0 \u00a0 if self.content:
\n\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 prompt += f"{self.content}<|eot_id|>"
\n\u00a0 \u00a0 \u00a0 \u00a0 return prompt<\/p>\n
class LlamaPrompt:
\n\u00a0 \u00a0 def __init__(self):
\n\u00a0 \u00a0 \u00a0 \u00a0 self.system_prompt = LlamaMessage("system", "You are a helpful assistant.")
\n\u00a0 \u00a0 \u00a0 \u00a0 self.history = History()<\/p>\n
\u00a0 \u00a0 def get_generation_prompt(self, tool_enabled=False, last=50):
\n\u00a0 \u00a0 \u00a0 \u00a0 prompt = [self.system_prompt] + self.history.get_chat_history(last) + [LlamaMessage("assistant")]
\n\u00a0 \u00a0 \u00a0 \u00a0 return ''.join([msg.template(tool_enabled) for msg in prompt])<\/p>\n
LlamaPrompt\u7c7b\u652f\u6301\u52a8\u6001\u6dfb\u52a0\u7528\u6237\u3001\u52a9\u624b\u3001\u5de5\u5177\u7ed3\u679c\u7b49\u89d2\u8272\u7684\u6d88\u606f，\u5e76\u6839\u636etool_enabled\u53c2\u6570\u51b3\u5b9a\u662f\u5426\u5728\u63d0\u793a\u4e2d\u5305\u542b\u5de5\u5177\u8c03\u7528\u8bf4\u660e，\u907f\u514d\u8f7b\u91cf\u7ea7\u6a21\u578b\u56e0\u6301\u7eed\u770b\u5230\u5de5\u5177\u6307\u4ee4\u800c\u4ea7\u751f\u6df7\u6dc6\u3002<\/p>\n
\u667a\u80fd\u4f53\u901a\u8fc7\u6b63\u5219\u8868\u8fbe\u5f0f\u5339\u914d\u5de5\u5177\u8c03\u7528\u6a21\u5f0f，\u89e3\u6790\u51fd\u6570\u540d\u4e0e\u53c2\u6570，\u5e76\u8c03\u7528\u5bf9\u5e94\u7684MCP\u5de5\u5177：<\/p>\n
class Agent:
\n\u00a0 \u00a0 def __init__(self, model, prompt, mcp_manager):
\n\u00a0 \u00a0 \u00a0 \u00a0 self.llm = model
\n\u00a0 \u00a0 \u00a0 \u00a0 self.prompt = prompt
\n\u00a0 \u00a0 \u00a0 \u00a0 self.mcp_manager = mcp_manager
\n\u00a0 \u00a0 \u00a0 \u00a0 self.tool_pattern = re.compile(r'\\\\[([A-Za-z0-9\\\\_]+\\\\(.*?\\\\),?\\\\s?)+\\\\]')<\/p>\n
\u00a0 \u00a0 def _is_tool_required(self, response):
\n\u00a0 \u00a0 \u00a0 \u00a0 return bool(self.tool_pattern.match(response))<\/p>\n
\u00a0 \u00a0 async def get_result_tool(self, response):
\n\u00a0 \u00a0 \u00a0 \u00a0 results = []
\n\u00a0 \u00a0 \u00a0 \u00a0 for name, params in self.parse_func_params(response):
\n\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 is_error, content = await self.mcp_manager.call_tool(name, params)
\n\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 results.append({"name": name, "output": [c.text for c in content]})
\n\u00a0 \u00a0 \u00a0 \u00a0 return json.dumps(results)<\/p>\n
\u667a\u80fd\u4f53\u7684chat\u65b9\u6cd5\u5b9e\u73b0\u5b8c\u6574\u7684\u5bf9\u8bdd\u6d41\u7a0b：<\/p>\n
\u5c06\u7528\u6237\u95ee\u9898\u4e0e\u5de5\u5177\u8c03\u7528\u8bf4\u660e\u7ec4\u5408\u4e3a\u7528\u6237\u63d0\u793a\u3002<\/p>\n<\/li>\n
\u751f\u6210\u521d\u6b65\u56de\u7b54，\u5224\u65ad\u662f\u5426\u9700\u8981\u5de5\u5177\u8c03\u7528\u3002<\/p>\n<\/li>\n
\u82e5\u9700\u8981，\u8c03\u7528MCP\u5de5\u5177\u5e76\u83b7\u53d6\u7ed3\u679c\u3002<\/p>\n<\/li>\n
\u7ed3\u679c\u5408\u6210：\u5c06\u5de5\u5177\u7ed3\u679c\u52a0\u5165\u63d0\u793a，\u751f\u6210\u6700\u7ec8\u56de\u7b54\u3002<\/p>\n
async def chat(self, question):
\n\u00a0tool_scheme = TOOL_CALL_PROMPT.format(function_scheme=self.mcp_manager.get_func_scheme())
\n\u00a0user_msg = self.prompt.get_user_prompt(question, tool_scheme)
\n\u00a0self.prompt.append_history(user_msg)<\/p>\n
\u00a0response = self.llm.generate(self.prompt.get_generation_prompt(tool_enabled=True))
\n\u00a0if self._is_tool_required(response):
\n\u00a0 \u00a0 \u00a0tool_result = await self.get_result_tool(response)
\n\u00a0 \u00a0 \u00a0tool_msg = self.prompt.get_tool_result_prompt(tool_result)
\n\u00a0 \u00a0 \u00a0self.prompt.append_history(tool_msg)
\n\u00a0 \u00a0 \u00a0response = self.llm.generate(self.prompt.get_generation_prompt(tool_enabled=False))<\/p>\n
\u00a0return response\n <\/li>\n
\u901a\u8fc7tool_enabled\u53c2\u6570\u7684\u5207\u6362，\u667a\u80fd\u4f53\u5728\u5de5\u5177\u8c03\u7528\u51b3\u7b56\u9636\u6bb5\u5305\u542b\u5de5\u5177\u6307\u4ee4，\u800c\u5728\u7ed3\u679c\u5408\u6210\u9636\u6bb5\u79fb\u9664\u6307\u4ee4，\u907f\u514d\u6a21\u578b\u8fc7\u5ea6\u5173\u6ce8\u5de5\u5177\u8c03\u7528，\u63d0\u5347\u56de\u7b54\u7684\u8fde\u8d2f\u6027\u3002<\/p>\n
\u5408\u6210\u56de\u7b54\u65f6\u51fa\u73b0\u7a7a\u54cd\u5e94\u6216\u65e0\u5173\u5de5\u5177\u8c03\u7528\u3002<\/p>\n<\/li>\n
\u6a21\u578b\u8fc7\u5ea6\u4f9d\u8d56\u5de5\u5177，\u5373\u4f7f\u5df2\u6709\u7ed3\u679c\u4ecd\u91cd\u590d\u8c03\u7528\u3002<\/p>\n<\/li>\n
\u7528\u6237\u67e5\u8be2\u7279\u5b9a\u77e5\u8bc6\u5185\u5bb9\u540e，\u6a21\u578b\u5728\u56de\u7b54\u65f6\u9519\u8bef\u8c03\u7528list_knowledges\u5de5\u5177\u3002<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n
\u4ec5\u5728\u751f\u6210\u5de5\u5177\u8c03\u7528\u51b3\u7b56\u65f6\u5305\u542b\u5de5\u5177\u6307\u4ee4\u3002<\/p>\n<\/li>\n
\u7ed3\u679c\u5408\u6210\u9636\u6bb5\u79fb\u9664\u6307\u4ee4，\u4e13\u6ce8\u4e8e\u5185\u5bb9\u6574\u5408\u3002<\/p>\n<\/li>\n<\/ul>\n<\/li>\n
\u56de\u7b54\u76f8\u5173\u6027\u663e\u8457\u63d0\u9ad8，\u5de5\u5177\u8c03\u7528\u66f4\u7cbe\u51c6\u3002<\/p>\n<\/li>\n
\u6a21\u578b\u80fd\u6709\u6548\u5229\u7528\u5de5\u5177\u7ed3\u679c，\u5982\u6b63\u786e\u89e3\u6790\u77e5\u8bc6\u5185\u5bb9\u5e76\u751f\u6210\u6458\u8981\u3002<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n
\u201c\u8bf7\u603b\u7ed3Obsidian\u77e5\u8bc6\u5e93\u4e2d\u5173\u4e8e\u2018AI\u4f26\u7406\u2019\u7684\u7b14\u8bb0，\u5e76\u4ee5Markdown\u8868\u683c\u5448\u73b0\u3002\u201d<\/p>\n<\/li>\n
\u6a21\u578b\u8c03\u7528get_knowledge_by_uri\u5de5\u5177\u83b7\u53d6\u7b14\u8bb0\u5185\u5bb9\u3002<\/p>\n<\/li>\n
\u6839\u636e\u5185\u5bb9\u957f\u5ea6\u81ea\u52a8\u683c\u5f0f\u5316\u4e3a\u8868\u683c，\u5c3d\u7ba1\u5b58\u5728\u8f7b\u5fae\u53c2\u6570\u9519\u8bef，\u4f46\u7ed3\u679c\u7ed3\u6784\u5316\u7a0b\u5ea6\u9ad8\u3002<\/p>\n<\/li>\n<\/li>\n<\/ul>\n
\u201c\u5217\u51fa\u6807\u9898\u5b58\u5728\u4f46\u5185\u5bb9\u4e3a\u7a7a\u7684\u7b14\u8bb0\u3002\u201d<\/p>\n<\/li>\n
\u6a21\u578b\u8c03\u7528list_knowledges\u83b7\u53d6\u6240\u6709\u7b14\u8bb0\u5143\u6570\u636e\u3002<\/p>\n<\/li>\n
\u901a\u8fc7\u6587\u4ef6\u5b57\u8282\u5927\u5c0f\u5224\u65ad\u7a7a\u7b14\u8bb0，\u7ed3\u679c\u90e8\u5206\u6b63\u786e，\u9700\u8fdb\u4e00\u6b65\u4f18\u5316\u7b5b\u9009\u903b\u8f91\u3002<\/p>\n<\/li>\n<\/li>\n<\/ul>\n
\u201c\u6839\u636e\u2018\u673a\u5668\u5b66\u4e60\u57fa\u7840\u2019\u7b14\u8bb0\u5185\u5bb9，\u751f\u62105\u4e2a\u7b80\u7b54\u9898\u3002\u201d<\/p>\n<\/li>\n
\u8c03\u7528\u5de5\u5177\u83b7\u53d6\u7b14\u8bb0\u5185\u5bb9\u3002<\/p>\n<\/li>\n
\u6a21\u578b\u5206\u6790\u5185\u5bb9\u7ed3\u6784，\u751f\u6210\u7b26\u5408\u8981\u6c42\u7684\u95ee\u9898，\u5982\u201c\u4ec0\u4e48\u662f\u76d1\u7763\u5b66\u4e60？\u201d<\/p>\n<\/li>\n<\/li>\n<\/ul>\n
Llama 3.2 3B\u6a21\u578b\u5728\u672c\u5730CPU（MacBook M1）\u4e0a\u7684\u63a8\u7406\u901f\u5ea6\u7ea6\u4e3a15 tokens\/\u79d2，\u9002\u5408\u4ea4\u4e92\u5f0f\u573a\u666f\u3002<\/p>\n<\/li>\n
\u591a\u8f6e\u5de5\u5177\u8c03\u7528\u80fd\u529b\u8f83\u5f31，\u96be\u4ee5\u5904\u7406\u590d\u6742\u63a8\u7406\u4efb\u52a1\u3002<\/p>\n<\/li>\n
\u5bf9\u6a21\u7cca\u67e5\u8be2\u7684\u7406\u89e3\u4e0d\u8db3，\u9700\u660e\u786e\u53c2\u6570（\u5982\u6b63\u786e\u7684URI）\u624d\u80fd\u6709\u6548\u6267\u884c\u5de5\u5177\u3002<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n
\u4ee3\u7801\u4ed3\u5e93：GitHub – hjlee94\/mcp-knowledge-base: MCP agent\/client\/server implementation for private knowledge base<\/p>\n","protected":false},"excerpt":{"rendered":"
\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb962\u6b21\uff0c\u70b9\u8d5e25\u6b21\uff0c\u6536\u85cf41\u6b21\u3002\u672c\u6587\u63a2\u8ba8\u4e86\u5982\u4f55\u6784\u5efa\u4e00\u4e2a\u5b8c\u5168\u672c\u5730\u5316\u7684AI\u667a\u80fd\u4f53\uff0c\u4ee5\u89e3\u51b3\u9690\u79c1\u4fdd\u62a4\u548c\u6570\u636e\u4e3b\u6743\u95ee\u9898\u3002\u901a\u8fc7\u81ea\u5b9a\u4e49\u7684ModelContextProtocol\uff08MCP\uff09\u670d\u52a1\u5668\u548cLlama3.2\u8f7b\u91cf\u7ea7\u6a21\u578b\uff0c\u5b9e\u73b0\u4e86\u77e5\u8bc6\u9694\u79bb\u548c\u5de5\u5177\u8c03\u7528\u80fd\u529b\u3002MCP\u670d\u52a1\u5668\u901a\u8fc7\u53ea\u8bfb\u8bbf\u95ee\u63a7\u5236\u3001\u8def\u5f84\u9690\u79c1\u4fdd\u62a4\u548c\u534f\u8bae\u6df1\u5ea6\u7406\u89e3\uff0c\u786e\u4fdd\u6570\u636e\u5728\u672c\u5730\u8fd0\u884c\uff0c\u907f\u514d\u6cc4\u9732\u98ce\u9669\u3002Llama3.2\u6a21\u578b\u56e0\u5176\u8bbe\u5907\u53cb\u597d\u6027\u3001\u5de5\u5177\u8c03\u7528\u652f\u6301\u548c\u6027\u80fd\u5e73\u8861\uff0c\u6210\u4e3a\u672c\u5730\u5316\u667a\u80fd\u4f53\u7684\u7406\u60f3\u9009\u62e9\u3002\u667a\u80fd\u4f53\u67b6\u6784\u5305\u62ecMCP\u5ba2\u6237\u7aef\u4e0e\u7ba1\u7406\u5668\u3001LLM\u6a21\u578b\u63a5\u53e3\u548c\u667a\u80fd\u4f53\u903b\u8f91\u5c42\uff0c\u901a\u8fc7\u5de5\u5177\u8c03\u7528\u51b3\u7b56\u4e0e\u7ed3\u679c\u5904\u7406\uff0c\u5b9e\u73b0\u95ee\u7b54\u903b\u8f91\u3002\u5b9e\u9a8c\u8868\u660e\uff0c\u8be5\u65b9\u6848\u5728\u77e5<\/p>\n","protected":false},"author":2,"featured_media":39122,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[347,73,3152],"topic":[],"class_list":["post-39123","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-llama","tag-mcp","tag-llm"],"yoast_head":"\n