A step-by-step guide to connecting Claude Desktop or Claude Web to your Postgres, Snowflake, or other internal databases using MarcoPolo MCP — no custom code required.
Before you start, make sure you have: a MarcoPolo account (free tier works), a paid Claude plan, and credentials for the database you want to connect. That’s it. No servers to provision, no SDKs to install, no infrastructure to manage.
The first step depends on which Claude client you’re using. For Claude Web, go to Settings → Connectors → Add Connector and paste the MarcoPolo MCP URL: https://mcp.marcopolo.dev. Authorize with your Google or GitHub account. You’re connected.
For Claude Desktop, download the MarcoPolo extension from the MarcoPolo dashboard and double-click the .mcpb file to install. Works on macOS, Windows, and Linux. For Claude Code, run: claude mcp add marcopolo --transport http https://mcp.marcopolo.dev then authenticate using /mcp => marcopolo => Authenticate.
Once MarcoPolo is connected to your Claude client, open the MarcoPolo dashboard. Navigate to Data Sources and click Add Source. Select your database type — Postgres, Snowflake, Redshift, BigQuery, and 47+ others are supported. Enter your connection credentials. MarcoPolo validates the connection and automatically explores your schema: tables, columns, relationships, data types.
This schema discovery is important. MarcoPolo doesn’t just store your credentials — it builds a semantic understanding of your data structure so your AI can reason about it effectively without you needing to explain it every time.
That’s genuinely it. Go back to Claude and start talking to your data. You can ask things like: “What were our top 10 customers by revenue last quarter?” or “Show me all orders that shipped more than 5 days late in January” or “Are there any users who signed up in the last 30 days but never completed their first action?”
Claude, via MarcoPolo, will generate the appropriate SQL, validate it against your schema, execute it in an isolated sandbox, and return the results — with an explanation of what it did and why.
When you ask a question, MarcoPolo’s runtime intercepts the tool call from Claude, injects the relevant schema context (without loading your entire database into the prompt), generates and validates the query, executes it in an isolated Kubernetes environment, and returns structured results.
The key insight is that MarcoPolo doesn’t just pass your question to the database. It manages the entire execution pipeline — context, validation, isolation, and results — so Claude can reason effectively without ever having direct database access.
Once you’ve connected your first data source, try adding a second one. The real power of MarcoPolo shows when Claude can reason across multiple systems in a single workflow — joining data from your warehouse with records from your CRM without you having to do anything differently.
