AnalyticDB PostgreSQL MCP Server

What is AnalyticDB PostgreSQL MCP Server about?

The server acts as a bridge between AI agents and AnalyticDB PostgreSQL databases, allowing agents to fetch schema information, run SELECT/DML/DDL statements, obtain execution plans, and interact with integrated GraphRAG and LLM memory services.

How to use AnalyticDB PostgreSQL MCP Server?

1. Install – either clone the repository and run with uv, or install the PyPI package adbpg_mcp_server via pip install adbpg_mcp_server.
2. Configure – add an entry to the MCP client configuration file under mcpServers with the appropriate command, arguments, and required environment variables (database connection details, GraphRAG keys, LLM‑memory keys).
3. Run – start the server (uv run adbpg-mcp-server or uvx adbpg_mcp_server).
4. Consume – AI agents can now call the exposed tools (SQL execution, table analysis, graphRAG upload/query, memory add/get/search/delete) and resources (schema, tables, DDL, statistics) through the MCP protocol.

Key Features

• Execute SELECT, DML, and DDL statements on AnalyticDB PostgreSQL.
• Retrieve database metadata: schemas, tables, DDL, statistics.
• Explain query execution plans.
• GraphRAG integration: upload documents, query knowledge graphs, manage decision trees.
• LLM long‑term memory API: add, retrieve, search, and delete memories linked to users, runs, or agents.
• Simple configuration via environment variables.
• Supports both source‑code deployment (uv) and pip‑installed package (uvx).

Use Cases

• AI‑driven data assistants that need to explore database schemas and run ad‑hoc queries.
• Knowledge‑graph augmentation where documents are uploaded to GraphRAG and later queried by agents.
• Long‑term conversational memory for agents that require context persistence across sessions.
• Automated ETL/analysis pipelines that trigger DDL/DML operations based on AI reasoning.
• Interactive analytics dashboards powered by AI agents that fetch statistics and execution plans on demand.

FAQ

• Do I need a specific Python version? Yes, Python 3.10 or higher.
• Which packages are required? mcp>=1.4.0, psycopg>=3.1.0, python-dotenv>=1.0.0, pydantic>=2.0.0.
• Can I run the server without GraphRAG or LLM memory? Yes, only set the required database env vars; the optional GraphRAG/LLMEMORY vars can be omitted if those features are not needed.
• How do I secure the credentials? Use a secrets manager or environment‑variable injection; never commit raw passwords to source control.
• What command should I use in the MCP client config? The example uses uv (or uvx for the pip package). Adjust the command and args accordingly.

AnalyticDB PostgreSQL MCP Server

AnalyticDB PostgreSQL MCP Server serves as a universal interface between AI Agents and AnalyticDB PostgreSQL databases. It enables seamless communication between AI Agents and AnalyticDB PostgreSQL, helping AI Agents retrieve database metadata and execute SQL operations.

Installation

You can set up the server either from the source code for development or by installing it from PyPI for direct use.

Option 1: From Source (for Development)

This method is recommended if you want to modify or contribute to the server.

# 1. Clone the repository
git clone https://github.com/aliyun/alibabacloud-adbpg-mcp-server.git
cd alibabacloud-adbpg-mcp-server

# 2. Create and activate a virtual environment using uv
uv venv .venv
source .venv/bin/activate  # On Linux/macOS
# .\.venv\Scripts\activate  # On Windows

# 3. Install the project in editable mode
uv pip install -e .

Option 2: From PyPI (for Production/Usage)

This is the simplest way to install the server for direct use within your projects.

pip install adbpg-mcp-server

Running the Server

The server can be run in two transport modes: stdio (default) for integration with MCP clients, and http for direct API access or debugging.

Make sure you have set up the required Environment Variables before running the server.

Stdio Mode (Default)

This is the standard mode for communication with an MCP client.

# Run using the default transport (stdio)
uv run adbpg-mcp-server

# Or explicitly specify the transport
uv run adbpg-mcp-server --transport stdio

Streamable-HTTP Mode

This mode exposes an HTTP server, which is useful for testing, debugging, or direct integration via REST APIs.

# Run the server in HTTP mode on the default host and port (127.0.0.1:3000)
uv run adbpg-mcp-server --transport http

# Specify a custom host and port
uv run adbpg-mcp-server --transport http --host 0.0.0.0 --port 3000

MCP Integration

To integrate this server with a parent MCP client, add the following configuration to the client's configuration file. The arguments in the args array will depend on the transport protocol you choose.

Example for Stdio Transport

"mcpServers": {
  "adbpg-mcp-server": {
    "command": "uv",
    "args": [
      "run",
      "adbpg-mcp-server",
      "--transport",
      "stdio"
    ],
    "env": {
      "ADBPG_HOST": "host",
      "ADBPG_PORT": "port",
      "ADBPG_USER": "username",
      "ADBPG_PASSWORD": "password",
      "ADBPG_DATABASE": "database",
      "GRAPHRAG_API_KEY": "graphrag llm api key",
      "GRAPHRAG_BASE_URL": "graphrag llm base url",
      "GRAPHRAG_LLM_MODEL": "graphrag llm model name",
      "GRAPHRAG_EMBEDDING_MODEL": "graphrag embedding model name",
      "GRAPHRAG_EMBEDDING_API_KEY": "graphrag embedding api key",
      "GRAPHRAG_EMBEDDING_BASE_URL": "graphrag embedding url",
      "LLMEMORY_API_KEY": "llm memory api_key",
      "LLMEMORY_BASE_URL": "llm memory base_url",
      "LLMEMORY_LLM_MODEL": "llm memory model name",
      "LLMEMORY_EMBEDDING_MODEL": "llm memory embedding model name",
      "LLMEMORY_ENABLE_GRAPH": "enable graph engine for llm memory (Default: false)"
    }
  }
}

Note: Since stdio is the default, you can optionally omit "--transport", "stdio" from the args array.

Example for Streamable-HTTP Transport

"mcpServers": {
  "adbpg-mcp-server": {
    "command": "uv",
    "args": [
      "run",
      "adbpg-mcp-server",
      "--transport",
      "http",
      "--port",
      "3000"
    ],
    "env": {
      "ADBPG_HOST": "host",
      "ADBPG_PORT": "port",
      "ADBPG_USER": "username",
      "ADBPG_PASSWORD": "password",
      "ADBPG_DATABASE": "database",
      "GRAPHRAG_API_KEY": "graphrag llm api key",
      "GRAPHRAG_BASE_URL": "graphrag llm base url",
      "GRAPHRAG_LLM_MODEL": "graphrag llm model name",
      "GRAPHRAG_EMBEDDING_MODEL": "graphrag embedding model name",
      "GRAPHRAG_EMBEDDING_API_KEY": "graphrag embedding api key",
      "GRAPHRAG_EMBEDDING_BASE_URL": "graphrag embedding url",
      "LLMEMORY_API_KEY": "llm memory api_key",
      "LLMEMORY_BASE_URL": "llm memory base_url",
      "LLMEMORY_LLM_MODEL": "llm memory model name",
      "LLMEMORY_EMBEDDING_MODEL": "llm memory embedding model name",
      "LLMEMORY_ENABLE_GRAPH": "enable graph engine for llm memory (Default: false)"
    }
  }
}

Tools

• execute_select_sql: Execute SELECT SQL queries on the AnalyticDB PostgreSQL server

• execute_dml_sql: Execute DML (INSERT, UPDATE, DELETE) SQL queries on the AnalyticDB PostgreSQL server

• execute_ddl_sql: Execute DDL (CREATE, ALTER, DROP) SQL queries on the AnalyticDB PostgreSQL server

• analyze_table: Collect table statistics

• explain_query: Get query execution plan

• adbpg_graphrag_upload

  • Description: Upload a text file (with its name) and file content to graphrag to generate a knowledge graph.
  • Parameters:
    • filename (text): The name of the file to be uploaded.
    • context (text): The textual content of the file.

• adbpg_graphrag_query

  • Description: Query the graphrag using the specified query string and mode。
  • Parameters:
    • query_str (text): the query content.
    • query_mode (text): The query mode, choose from [bypass, naive, local, global, hybrid, mix]. If null, defaults to mix.

• adbpg_graphrag.upload_decision_tree(context text, root_node text)

  • Description: Upload a decision tree with the specified root_node. If the root_node does not exist, a new decision tree will be created.
  • Parameters:
    • context (text): The textual representation of the decision tree.
    • root_node (text): The content of the root node.

• adbpg_graphrag.append_decision_tree(context text, root_node_id text)

  • Description: Append a subtree to an existing decision tree at the node specified by root_node_id.
  • Parameters:
    • context (text): The textual representation of the subtree.
    • root_node_id (text): The ID of the node to which the subtree will be appended.

• adbpg_graphrag.delete_decision_tree(root_node_entity text)

  • Description: Delete a sub-decision tree under the node specified by root_node_entity.
  • Parameters:
    • root_node_entity (text): The ID of the root node of the sub-decision tree to be deleted.

• adbpg_llm_memory_add

  • Description: Add LLM long memory.
  • Parameters:
    • messages (json): The name of the file to be uploaded.
    • user_id (text): The user id.
    • run_id (text): The run id.
    • agent_id (text): The agent id.
    • metadata (json): The metadata json(optional).
    • memory_type (text): The memory type(optional).
    • prompt (text): The prompt(optional). Note:
      At least one of user_id, run_id, or agent_id should be provided.

• adbpg_llm_memory_get_all

  • Description: Retrieves all memory records associated with a specific user, run or agent.
  • Parameters:
    • user_id (text): User ID (optional). If provided, fetch all memories for this user.
    • run_id (text): Run ID (optional).
    • agent_id (text): Agent ID (optional). If provided, fetch all memories for this agent. Note:
      At least one of user_id, run_id, or agent_id should be provided.

• adbpg_llm_memory_search

  • Description: Retrieves memories relevant to the given query for a specific user, run, or agent.
  • Parameters:
    • query (text): The search query string.
    • user_id (text): User ID (optional). If provided, fetch all memories for this user.
    • run_id (text): Run ID (optional).
    • agent_id (text): Agent ID (optional). If provided, fetch all memories for this agent.
    • filter (json): Additional filter conditions in JSON format (optional). Note:
      At least one of user_id, run_id, or agent_id should be provided.

• adbpg_llm_memory_delete_all:

  • Description: Delete all memory records associated with a specific user, run or agent.
  • Parameters:
    • user_id (text): User ID (optional). If provided, fetch all memories for this user.
    • run_id (text): Run ID (optional).
    • agent_id (text): Agent ID (optional). If provided, fetch all memories for this agent. Note:
      At least one of user_id, run_id, or agent_id should be provided.

Resources

Built-in Resources

• adbpg:///schemas: Get all schemas in the database

Resource Templates

• adbpg:///{schema}/tables: List all tables in a specific schema
• adbpg:///{schema}/{table}/ddl: Get table DDL
• adbpg:///{schema}/{table}/statistics: Show table statistics

Environment Variables

MCP Server requires the following environment variables to connect to AnalyticDB PostgreSQL instance:

• ADBPG_HOST: Database host address
• ADBPG_PORT: Database port
• ADBPG_USER: Database username
• ADBPG_PASSWORD: Database password
• ADBPG_DATABASE: Database name

MCP Server requires the following environment variables to initialize graphRAG and llm memory server：

• API_KEY: API key for LLM provider or embedding API
• BASE_URL: Base URL for LLM or embedding service endpoint
• LLM_MODEL: LLM model name or identifier
• EMBEDDING_MODEL: Embedding model name or identifier

Dependencies

• Python 3.11 or higher
• uv (for environment and package management)