Build a PostgreSQL MCP Server with n8n Workflow Automation

Build a PostgreSQL MCP server inside n8n—so an MCP client can safely run SELECT, INSERT, and UPDATE operations

This n8n automation template demonstrates how to build a simple PostgreSQL MCP server that lets a compatible MCP client manage your PostgreSQL database—covering common business datasets like HR, Payroll, Sales, Inventory, and more—using a structured tool interface rather than raw SQL.

What this workflow does

• Starts with an MCP server trigger and connects it to 5 tools: 2 PostgreSQL read-only tools and 3 custom workflow tools.
• Routes database operations through a switch that sends each request to the correct handler.
• Uses the 2 PostgreSQL tools for simple read-only queries (straightforward “use it as-is” queries).
• Uses the 3 custom workflow tools for select, insert, and update operations where extra discretion is required.
• Improves safety by accepting parameters via a restricted schema, then constructing the SQL statement internally—rather than allowing free-form raw SQL.
• Executes the correct operation using the workflow’s standard PostgreSQL node and returns the results back to the MCP client.

Use cases

• HR/Payroll lookups: retrieve records with MCP-driven read-only queries.
• Inventory updates: insert or update stock movement or item records via parameterized tool calls.
• SaaS operator workflows: integrate an MCP client to safely manage operational data without exposing raw SQL.
• Automation engineering: provide a clear pattern for “agent tools” backed by PostgreSQL in n8n.

Technical details

• MCP Trigger: “n8nn8n-nodes-langchainmcp trigger” (MCP server trigger)
• Routing: switch to route operations to the proper handler
• Database: postgres node for handling select/insert/update
• Supporting tools: “postgres tool” and the template’s custom workflow tools (select, insert, update) executed via an “Execute workflow” trigger

This MCP example is based on the official reference implementation: modelcontextprotocol/servers/src/postgres.