Warehouse Mcp
Let your coding agent run governed SQL against Postgres, Snowflake, BigQuery, and other warehouses from one MCP server.
Overview
warehouse-mcp is a MCP server for the Build phase that connects agents to Postgres, Oracle, Snowflake, BigQuery, Redshift, DuckDB, and MotherDuck for governed warehouse queries.
What is this MCP server?
- warehouse-mcp npm v0.4.2 with required start positional argument
- Seven warehouse engines: postgres, oracle, redshift, snowflake, bigquery, duckdb (MotherDuck via DUCKDB_PATH=md:<db>)
- stdio for Claude Desktop and Cursor via MCP_TRANSPORT=stdio
- init and doctor CLI entry paths for setup and configuration diagnosis
- Server version 0.4.2
- 6 explicit WAREHOUSE_TYPE choices plus MotherDuck via DuckDB path convention
- 3 CLI modes documented: start, init, doctor
What problem does it solve?
Copy-pasting SQL results between a warehouse console and your AI agent breaks flow and encourages unsafe, unscoped database access.
Who is it for?
Indie SaaS founders who already have a cloud or DuckDB warehouse and want agent-native analytics during build and growth.
Skip if: Greenfield projects with no database yet, or teams forbidden from exposing warehouse credentials to desktop MCP clients.
What do I get? / Deliverables
After configuring WAREHOUSE_TYPE, transport, and API keys, your agent queries the live warehouse through MCP tools with a single npm-managed server process.
- Live MCP SQL access to the configured warehouse engine
- Diagnosed connection config via doctor when misconfigured
- Repeatable stdio or HTTP MCP endpoint for analytics questions
Recommended MCP Servers
Journey fit
Build is the primary shelf because warehouse MCP is most often wired while integrating data backends and analytics into the product. Integrations subphase reflects multi-engine connection setup, WAREHOUSE_TYPE selection, and MCP API key gated access.
How it compares
Multi-engine warehouse MCP, not a single-vendor BI skill or an ETL orchestration playbook.
Common Questions / FAQ
Who is warehouse-mcp for?
warehouse-mcp is for developers using Claude Desktop, Cursor, or similar MCP clients who need one server package to query Postgres, Snowflake, BigQuery, Redshift, Oracle, DuckDB, or MotherDuck.
When should I use warehouse-mcp?
Use it when the product or ops stack already stores data in a supported warehouse and you want the agent to explore metrics or debug data without leaving the IDE.
How do I add warehouse-mcp to my agent?
Install the warehouse-mcp npm package, run with the required start argument, set WAREHOUSE_TYPE and credentials, set MCP_TRANSPORT=stdio for desktop clients, and register the stdio server in your MCP config.