What it does
Keboola MCP Server bridges Keboola—a data integration and orchestration platform—to AI assistants including Claude, Cursor, and Windsurf. It exposes Keboola's core features as callable tools: run SQL queries and transformations, trigger and monitor job execution, query and manage storage tables, create and deploy data apps, build workflow pipelines, and search project metadata. The server supports both remote (OAuth-backed, hosted on Keboola's infrastructure) and local deployment modes, and respects development branches for safe testing outside production.
Who it's for
Data engineers and AI developers using Keboola for ETL pipelines who want to automate or interact with data workflows through AI agents. Useful for teams building internal AI tools, analytics automation, or agentic data operations where Claude or Cursor serve as the primary control plane for Keboola workflows.
Common use cases
- Query Keboola storage tables and run analytics directly from Claude Code without leaving your IDE.
- Create SQL transformations and manage extractor or writer components using natural language prompts.
- Trigger data pipeline jobs and monitor execution status programmatically via an AI agent.
- Build and deploy Streamlit data apps connected to Keboola storage tables.
- Search and update project metadata and documentation at scale.
Setup pitfalls
- One secret was detected in the repository—manage credentials via environment variables, never commit them to version control.
- CI tests are currently failing; verify the latest release in a non-production environment before deploying.
- OAuth authentication is required; ensure your Keboola account and project are properly configured before connecting.
- The server reads and writes to the filesystem and makes network calls to Keboola—sandbox or restrict accordingly in your AI client.