Keboola MCP Server

What is Keboola MCP Server?

Keboola MCP Server is an open-source bridge between a Keboola project and AI tools such as Claude, Cursor, CrewAI, LangChain, and Amazon Q. It exposes Keboola features—storage access, SQL transformations, job triggers, and more—as callable tools, so AI agents can read and write data without glue code.

How to use Keboola MCP Server?

The easiest way is the remote hosted server: navigate to your Keboola Project Settings → MCP Server tab, copy the server URL (e.g., https://mcp.<YOUR_REGION>.keboola.com/mcp), and paste it into your MCP client. For local use, clone the repo, set KBC_STORAGE_TOKEN, KBC_STORAGE_API_URL, and optionally KBC_WORKSPACE_SCHEMA and KBC_BRANCH_ID, then start the server with uv run mcp-keboola --transport stdio or --transport streamable-http.

Key features of Keboola MCP Server

• Query tables and manage table or bucket descriptions
• Create, list, and inspect extractors, writers, and transformation configurations
• Create SQL transformations using natural language
• Run components, transformations, and retrieve job details
• Build workflow pipelines with Conditional Flows and Orchestrator Flows
• Create, deploy, and manage Keboola Streamlit Data Apps

Use cases of Keboola MCP Server

• Let Claude or Cursor browse project tables and schemas for context-aware data work
• Trigger a transformation or extraction job directly from a chat conversation
• Query storage tables and visualize results in a Data App without writing integration code
• Safely prototype new pipelines in a development branch before promoting to production

FAQ from Keboola MCP Server

What do I need to use the remote server?

You need a Keboola project on a multi-tenant stack and an AI assistant that supports remote Streamable HTTP connections with OAuth. The server URL is available in Project Settings → MCP Server.

Can I restrict which tools an AI agent can use?

Yes. When using HTTP transports, you can set X-Allowed-Tools, X-Disallowed-Tools, or X-Read-Only-Mode headers to limit available tools. Read‑only tools only retrieve data without modifying the project.

How do I work on a development branch instead of production?

Set the KBC_BRANCH_ID environment variable (or the X-Branch-Id header on remote transports) to the numeric branch ID. All operations are then scoped to that branch. If omitted, the production branch is used.

Do I need to install Python or any dependencies?

The remote server requires no installation on your machine. For the local setup you need Python 3.10+, uv, and a Keboola project with admin rights.

What transports does the local server support?

It supports stdio (default, single client over standard I/O), streamable-http (bidirectional HTTP streaming at /mcp), and http-compat (alias for streamable-http).