基于AnalyticDB PostgreSQL MCP Server实现RAG与长记忆

版本限制

内核版本为7.2.1.3及以上的AnalyticDB for PostgreSQL7.0版实例。

前提条件

• 如果您需要使用长记忆功能，请提交工单联系技术支持协助安装adbpg_llm_memory插件。

• 获取调用大模型服务的API Key。如果您通过阿里云百炼平台调用大模型，获取方法请参见API Key。

操作步骤

本章节以MCP客户端调用MCP Server功能为例，介绍安装与调用MCP Server的操作过程。

import asyncio
import json

from mcp.client.stdio import stdio_client
from mcp import ClientSession, StdioServerParameters

config_file_path = "config.json"

# 读取并解析 JSON 配置文件
try:
    with open(config_file_path, "r") as f:
        config_data = json.load(f)
    # 从解析后的数据中提取我们需要的服务器配置
    server_config = config_data["mcpServers"]["adbpg-mcp-server"]
except FileNotFoundError:
    print(f"错误：配置文件 '{config_file_path}' 未找到。请确保文件存在于当前目录。")
    exit(1)
except (KeyError, TypeError):
    print(f"错误：配置文件 '{config_file_path}' 格式不正确。请检查其结构。")
    exit(1)

# 为 stdio 连接创建服务器参数
server_params = StdioServerParameters(
    # 服务器执行的命令，使用 uv 运行 adbpg_mcp_server.py
    command = server_config.get("command"),
    # 运行的参数
    args = server_config.get("args"),
    # 环境变量，默认为 None，表示使用当前环境变量
    env = server_config.get("env")
)

async def main():
    # 创建 stdio 客户端
    async with stdio_client(server_params) as (stdio, write):
        # 创建 ClientSession 对象
        async with ClientSession(stdio, write) as session:
            # 初始化 ClientSession
            await session.initialize()

            # 列出可用的工具
            response = await session.list_tools()
            print(response)
            print("\n--------------------------------\n")

            # 调用工具 execute_select_sql
            response = await session.call_tool('execute_select_sql', {'query': 'select 1;'})
            print(response)
            print("\n--------------------------------\n")

            # 调用工具 adbpg_graphrag_query
            response = await session.call_tool('adbpg_graphrag_query', {'query_str': 'who are you?', 'query_mode': 'bypass'})
            print(response)
            print("\n--------------------------------\n")
            
            # 调用工具 adbpg_llm_memory_add
            messages = [
                {"role": "user", "content": "嗨, 我是张三。 我喜欢徒步，不喜欢剧烈的运动。"},
                {"role": "assistant", "content": "你好，张三！徒步是个很棒的爱好，我会记住你的喜好。如果你有任何徒步路线规划、装备推荐或沿途风景的问题，欢迎随时交流。"}
            ]
            response = await session.call_tool('adbpg_llm_memory_add', {
                'messages': messages,
                'user_id': 'test_u',
            })
            print(response)
            print("\n--------------------------------\n")

            # 调用工具 adbpg_llm_memory_get_all
            response = await session.call_tool('adbpg_llm_memory_get_all', {'user_id': 'test_u'})
            print(response)
            print("\n--------------------------------\n")

if __name__ == '__main__':
    asyncio.run(main())

返回示例

2025-09-16 20:40:36,995 - mcp.server.lowlevel.server - DEBUG - Initializing server 'adbpg-mcp-server'
2025-09-16 20:40:36,995 - adbpg-mcp-server - INFO - MCP server initialized
2025-09-16 20:40:36,995 - mcp.server.lowlevel.server - DEBUG - Registering handler for ListResourcesRequest
2025-09-16 20:40:36,995 - mcp.server.lowlevel.server - DEBUG - Registering handler for ListResourceTemplatesRequest
2025-09-16 20:40:36,995 - mcp.server.lowlevel.server - DEBUG - Registering handler for ReadResourceRequest
2025-09-16 20:40:36,995 - mcp.server.lowlevel.server - DEBUG - Registering handler for ListToolsRequest
2025-09-16 20:40:36,995 - mcp.server.lowlevel.server - DEBUG - Registering handler for CallToolRequest
2025-09-16 20:40:36,996 - asyncio - DEBUG - Using selector: KqueueSelector
2025-09-16 20:40:36,996 - adbpg_mcp_server.adbpg - INFO - Connecting to database for _conn...
2025-09-16 20:40:37,145 - adbpg_mcp_server.adbpg - INFO - New database connection established and initialized for _conn (id: 4395876432)
2025-09-16 20:40:37,157 - adbpg-mcp-server - INFO - Successfully connected to database.
2025-09-16 20:40:37,157 - adbpg_mcp_server.adbpg - INFO - Connecting to database for _graphrag_conn...
2025-09-16 20:40:37,189 - adbpg_mcp_server.adbpg - INFO - Running initializer for _graphrag_conn on new connection...
2025-09-16 20:40:38,067 - adbpg_mcp_server.adbpg - INFO - New database connection established and initialized for _graphrag_conn (id: 4395131408)
2025-09-16 20:40:38,067 - adbpg-mcp-server - INFO - GraphRAG initialized successfully.
2025-09-16 20:40:38,067 - adbpg_mcp_server.adbpg - INFO - Connecting to database for _llm_memory_conn...
2025-09-16 20:40:38,111 - adbpg_mcp_server.adbpg - INFO - Running initializer for _llm_memory_conn on new connection...
2025-09-16 20:40:39,054 - adbpg_mcp_server.adbpg - INFO - New database connection established and initialized for _llm_memory_conn (id: 4395130768)
2025-09-16 20:40:39,054 - adbpg-mcp-server - INFO - LLM Memory initialized successfully.
2025-09-16 20:40:39,054 - adbpg-mcp-server - INFO - Starting ADBPG MCP server...
2025-09-16 20:40:39,064 - adbpg-mcp-server - INFO - Running MCP server with stdio transport...
2025-09-16 20:40:39,075 - mcp.server.lowlevel.server - DEBUG - Received message: root=InitializedNotification(method='notifications/initialized', params=None, jsonrpc='2.0')
2025-09-16 20:40:39,075 - mcp.server.lowlevel.server - DEBUG - Received message: <mcp.shared.session.RequestResponder object at 0x10607d1d0>
2025-09-16 20:40:39,075 - mcp.server.lowlevel.server - INFO - Processing request of type ListToolsRequest
2025-09-16 20:40:39,075 - mcp.server.lowlevel.server - DEBUG - Dispatching request of type ListToolsRequest
2025-09-16 20:40:39,076 - mcp.server.lowlevel.server - DEBUG - Response sent
meta=None nextCursor=None tools=[Tool(name='execute_select_sql', title=None, description='Execute SELECT SQL to query data from ADBPG database. Returns data in JSON format.', inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'description': 'The (SELECT) SQL query to execute'}}, 'required': ['query']}, outputSchema=None, annotations=None, meta=None), Tool(name='execute_dml_sql', title=None, description='Execute (INSERT, UPDATE, DELETE) SQL to modify data in ADBPG database.', inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'description': 'The DML SQL query to execute'}}, 'required': ['query']}, outputSchema=None, annotations=None, meta=None), Tool(name='execute_ddl_sql', title=None, description='Execute (CREATE, ALTER, DROP) SQL statements to manage database objects.', inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'description': 'The DDL SQL query to execute'}}, 'required': ['query']}, outputSchema=None, annotations=None, meta=None), Tool(name='analyze_table', title=None, description='Execute ANALYZE command to collect table statistics.', inputSchema={'type': 'object', 'properties': {'schema': {'type': 'string'}, 'table': {'type': 'string'}}, 'required': ['schema', 'table']}, outputSchema=None, annotations=None, meta=None), Tool(name='explain_query', title=None, description='Get query execution plan.', inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'description': 'The SQL query to analyze'}}, 'required': ['query']}, outputSchema=None, annotations=None, meta=None), Tool(name='adbpg_graphrag_upload', title=None, description='Execute graphrag upload operation', inputSchema={'type': 'object', 'properties': {'filename': {'type': 'string', 'description': 'The file name need to upload'}, 'context': {'type': 'string', 'description': 'the context of your file'}}, 'required': ['filename', 'context']}, outputSchema=None, annotations=None, meta=None), Tool(name='adbpg_graphrag_query', title=None, description='Execute graphrag query operation', inputSchema={'type': 'object', 'properties': {'query_str': {'type': 'string', 'description': 'The query you want to ask'}, 'query_mode': {'type': 'string', 'description': 'The query mode you need to choose [ bypass,naive, local, global, hybrid, mix[default], tree ].'}, 'start_search_node_id': {'type': 'string', 'description': "If using 'tree' query mode, set the start node ID of tree."}}, 'required': ['query_str']}, outputSchema=None, annotations=None, meta=None), Tool(name='adbpg_graphrag_upload_decision_tree', title=None, description=' Upload a decision tree with the specified root_node. If the root_node does not exist, a new decision tree will be created. ', inputSchema={'type': 'object', 'properties': {'root_node': {'type': 'string', 'description': 'the root_noot (optional)'}, 'context': {'type': 'string', 'description': 'the context of decision'}}, 'required': ['context']}, outputSchema=None, annotations=None, meta=None), Tool(name='adbpg_graphrag_append_decision_tree', title=None, description='Append a subtree to an existing decision tree at the node specified by root_node_id. ', inputSchema={'type': 'object', 'properties': {'root_node_id': {'type': 'string', 'description': 'the root_noot_id'}, 'context': {'type': 'string', 'description': 'the context of decision'}}, 'required': ['context', 'root_node_id']}, outputSchema=None, annotations=None, meta=None), Tool(name='adbpg_graphrag_delete_decision_tree', title=None, description=' Delete a sub-decision tree under the node specified by root_node_entity. ', inputSchema={'type': 'object', 'properties': {'root_node_entity': {'type': 'string', 'description': 'the root_noot_entity'}}, 'required': ['root_node_entity']}, outputSchema=None, annotations=None, meta=None), Tool(name='adbpg_graphrag_reset_tree_query', title=None, description=' Reset the decision tree in the tree query mode', inputSchema={'type': 'object', 'required': []}, outputSchema=None, annotations=None, meta=None), Tool(name='adbpg_llm_memory_add', title=None, description='Execute llm_memory add operation', inputSchema={'type': 'object', 'properties': {'messages': {'type': 'array', 'items': {'type': 'object', 'properties': {'role': {'type': 'string'}, 'content': {'type': 'string'}}, 'required': ['role', 'content']}, 'description': 'List of messages objects (e.g., conversation history)'}, 'user_id': {'type': 'string', 'description': 'the user_id'}, 'run_id': {'type': 'string', 'description': 'the run_id'}, 'agent_id': {'type': 'string', 'description': 'the agent_id'}, 'metadata': {'type': 'object', 'description': 'the metatdata json'}, 'memory_type': {'type': 'string', 'description': 'the memory_type text'}, 'prompt': {'type': 'string', 'description': 'the prompt'}}, 'required': ['messages']}, outputSchema=None, annotations=None, meta=None), Tool(name='adbpg_llm_memory_get_all', title=None, description='Execute llm_memory get_all operation', inputSchema={'type': 'object', 'properties': {'user_id': {'type': 'string', 'description': 'The user_id'}, 'run_id': {'type': 'string', 'description': 'The run_id'}, 'agent_id': {'type': 'string', 'description': 'The agent_id'}}, 'required': []}, outputSchema=None, annotations=None, meta=None), Tool(name='adbpg_llm_memory_search', title=None, description='Execute llm_memory search operation', inputSchema={'type': 'object', 'properties': {'query': {'type': 'string', 'description': 'llm_memory relevant query'}, 'user_id': {'type': 'string', 'description': 'The search of user_id'}, 'run_id': {'type': 'string', 'description': 'The search of run_id'}, 'agent_id': {'type': 'string', 'description': 'The search of agent_id'}, 'filter': {'type': 'object', 'description': 'The search of filter'}}, 'required': ['query']}, outputSchema=None, annotations=None, meta=None), Tool(name='adbpg_llm_memory_delete_all', title=None, description='Execute llm_memory delete_all operation', inputSchema={'type': 'object', 'properties': {'user_id': {'type': 'string', 'description': 'The user_id'}, 'run_id': {'type': 'string', 'description': 'The run_id'}, 'agent_id': {'type': 'string', 'description': 'The agent_id'}}, 'required': []}, outputSchema=None, annotations=None, meta=None)]

--------------------------------

2025-09-16 20:40:39,079 - mcp.server.lowlevel.server - DEBUG - Received message: <mcp.shared.session.RequestResponder object at 0x10607d310>
2025-09-16 20:40:39,079 - mcp.server.lowlevel.server - INFO - Processing request of type CallToolRequest
2025-09-16 20:40:39,079 - mcp.server.lowlevel.server - DEBUG - Dispatching request of type CallToolRequest
2025-09-16 20:40:39,095 - mcp.server.lowlevel.server - DEBUG - Response sent
meta=None content=[TextContent(type='text', text='[\n  {\n    "?column?": 1\n  }\n]', annotations=None, meta=None)] structuredContent=None isError=False

--------------------------------

2025-09-16 20:40:39,096 - mcp.server.lowlevel.server - DEBUG - Received message: <mcp.shared.session.RequestResponder object at 0x105fbd5b0>
2025-09-16 20:40:39,096 - mcp.server.lowlevel.server - INFO - Processing request of type CallToolRequest
2025-09-16 20:40:39,096 - mcp.server.lowlevel.server - DEBUG - Dispatching request of type CallToolRequest
2025-09-16 20:40:42,482 - mcp.server.lowlevel.server - DEBUG - Response sent
meta=None content=[TextContent(type='text', text='I am Qwen, a large language model developed by Alibaba Cloud. I have the ability to answer questions, write stories, emails, scripts, and more, as well as perform logical reasoning and programming tasks. I support multiple languages and strive to provide accurate and helpful information. How can I assist you today?', annotations=None, meta=None)] structuredContent=None isError=False

--------------------------------

2025-09-16 20:40:42,486 - mcp.server.lowlevel.server - DEBUG - Received message: <mcp.shared.session.RequestResponder object at 0x105fbcd60>
2025-09-16 20:40:42,487 - mcp.server.lowlevel.server - INFO - Processing request of type CallToolRequest
2025-09-16 20:40:42,487 - mcp.server.lowlevel.server - DEBUG - Dispatching request of type CallToolRequest
2025-09-16 20:40:48,351 - mcp.server.lowlevel.server - DEBUG - Response sent
meta=None content=[TextContent(type='text', text="{'results': []}", annotations=None, meta=None)] structuredContent=None isError=False

--------------------------------

2025-09-16 20:40:48,354 - mcp.server.lowlevel.server - DEBUG - Received message: <mcp.shared.session.RequestResponder object at 0x106080830>
2025-09-16 20:40:48,354 - mcp.server.lowlevel.server - INFO - Processing request of type CallToolRequest
2025-09-16 20:40:48,354 - mcp.server.lowlevel.server - DEBUG - Dispatching request of type CallToolRequest
2025-09-16 20:40:48,370 - mcp.server.lowlevel.server - DEBUG - Response sent
meta=None content=[TextContent(type='text', text="{'results': [{'id': '72ede2e1-a1ac-435c-8b10-10ffdfb7****', 'memory': '不喜欢剧烈的运动', 'hash': 'b0d6c2f0e5df599e02d8906925cb****', 'metadata': None, 'created_at': '2025-09-15T18:53:22.594877+08:00', 'updated_at': None, 'user_id': 'test_u'}, {'id': 'b041adbb-e191-44e1-84a5-1b83cf85****', 'memory': '名字是张三', 'hash': 'c31d55d5c8a2e6dd7055e46ca861****', 'metadata': None, 'created_at': '2025-09-15T18:53:22.578566+08:00', 'updated_at': None, 'user_id': 'test_u'}, {'id': '99f94dd0-3e7a-471a-8d90-64b670e2****', 'memory': '喜欢徒步', 'hash': 'a6ccce390824f2a7bdcc1a38c400****', 'metadata': None, 'created_at': '2025-09-15T18:53:22.590296+08:00', 'updated_at': None, 'user_id': 'test_u'}]}", annotations=None, meta=None)] structuredContent=None isError=False

--------------------------------

2025-09-16 20:40:48,372 - adbpg_mcp_server.adbpg - INFO - All database connections closed.

MCP Server支持的工具