MCP-PostgreSQL-Ops

MCP-PostgreSQL-Ops is an MCP server that provides PostgreSQL database operations, monitoring, and maintenance recommendations via natural language queries. It supports PostgreSQL versions 12 through 17 and can leverage optional extensions like pg_stat_statements and pg_stat_monitor for deeper query analytics. Once the MCP server is running, you can connect it through the OpenWebUI and use the built-in tools to perform read-only analysis, performance monitoring, bloat detection, and maintenance recommendations across one or more PostgreSQL databases. The server exposes its capabilities via the MCPO API (documented in the OpenWebUI docs), allowing you to ask questions like: Which queries are slow? What is the current VACUUM effectiveness? How bloated are our tables across databases? The toolset is designed to be safe for production environments, operating in read-only mode unless you explicitly enable non-destructive maintenance aids through controlled queries. To use the tools, ensure your OpenWebUI is connected to the MCP server address (e.g., http://localhost:8003/postgresql-ops) and then navigate to available endpoints or the swagger docs to issue natural-language queries and view structured results.

Prerequisites:

• Python 3.8+ (preferably 3.10+)
• pip (Python package manager)
• Network access to the target PostgreSQL instance (and necessary credentials)

Step 1: Install the MCP PostgreSQL Ops package

• Option A: Install from PyPI

  pip install MCP-PostgreSQL-Ops
  

• Option B: Install from source (clone repository and install)

  git clone https://github.com/call518/MCP-PostgreSQL-Ops.git
  cd MCP-PostgreSQL-Ops
  pip install -e .
  

Step 2: Configure environment (example values shown; replace with your own)

• Create a run configuration file or export environment variables as needed. Example:

  export POSTGRES_HOST=host.docker.internal
  export POSTGRES_PORT=15432
  export POSTGRES_USER=postgres
  export POSTGRES_PASSWORD=changeme!@34
  export POSTGRES_DB=ecommerce
  

Step 3: Run the MCP server

• If using the recommended Python module entrypoint:

  python -m mcp_postgresql_ops
  

• Or run via your preferred invocation method if using a packaged entrypoint.

Step 4: Verify the server is up

• Check logs for readiness and access the OpenWebUI at the configured URL (e.g., http://localhost:3003/ or as documented by your deployment).

Tips and notes:

• The MCP server operates in a read-only mode for safety in production environments. Ensure your PostgreSQL permissions align with read-only monitoring tasks unless you explicitly enable maintenance-facing features.
• If you use your own PostgreSQL instance, set POSTGRES_HOST, POSTGRES_PORT, POSTGRES_USER, POSTGRES_PASSWORD, and POSTGRES_DB accordingly. The README’s quickstart shows a built-in testing container; adjust docker-compose or environment variables if you point to an external database.
• Extensions like pg_stat_statements and pg_stat_monitor enhance query analytics. Install them in your PostgreSQL server if you want richer insights.
• When deploying with OpenWebUI, ensure the MCP server address is correctly configured in the UI to enable tool registration and usage.
• For multi-database analysis, the MCP server will aggregate data across accessible databases, provided the connected user has privileges as needed.