Keboola MCP Server

Keboola is a managed data platform — extractors, writers, SQL transformations, orchestrated flows and a governed storage layer on top of Snowflake or BigQuery. This official MCP server turns that platform into tools an agent can call, so a model can go from "which tables do we have on orders?" to a written, saved and executed transformation without anyone hand-writing glue code.

The tool surface follows the platform's own objects. Storage tools query tables directly and read or update bucket and table descriptions. Component tools list and inspect extractors, writers, data apps and transformation configurations, and create new ones. SQL tools author transformations from a natural-language brief in the project's own dialect. Job tools run components and transformations and return execution detail, so a failed run can be diagnosed in the same conversation. Flow tools build and manage both Conditional Flows and Orchestrator Flows. Data App tools create, deploy and manage Streamlit apps over project data. Metadata tools search, read and update project documentation and object metadata. Crucially, dev-branch support scopes every operation to a development branch, so an agent can be let loose without touching production.

The easiest path is the hosted remote server: every multi-tenant Keboola stack runs one at https://mcp.<region>.keboola.com/mcp with OAuth, so there is nothing to install — copy the URL from Project Settings → MCP Server, authenticate in the browser, then pick the project in conversation. For local or development use, the Python package keboola-mcp-server runs over stdio against a storage API token.

Clients confirmed by the project include Claude Desktop and Claude Code, Cursor, Windsurf, VS Code and Make, plus agent frameworks such as CrewAI, LangChain and Amazon Q. MIT-licensed and developed in the open at keboola/mcp-server.