Can AI agents automate Keboola?

Keboola is automatable today through its hosted MCP server and the kbagent command-line interface. This record is the Keboola data platform. Fivetran, Airbyte, Snowflake, and dbt are separate products. The Storage API takes a project Storage API token in the X-StorageApi-Token header. Storage calls use form data, while Keboola says the rest of its API uses JSON. Table import and export use CSV. Official Storage clients exist for Python, R, and PHP. Tokens are created under Project Settings, API Tokens. A master token belongs to a project administrator and can create other tokens. The hosted MCP server is on each multi-tenant stack at https://mcp.<region>.keboola.com/mcp and uses OAuth. The project settings page shows the regional URL. Tools cover storage, components, SQL transformations, jobs, flows, data apps, metadata, and dev branches. A local server can instead use a browser login or a Storage API token. The Keboola product page says the MCP server is open source, OAuth-authenticated, and not billed as a separate product. Query and job usage is billed as normal Keboola usage. kbagent is the current command-line interface. It prints JSON, logs in with kbagent auth login, and manages projects, configs, jobs, storage, flows, and dev branches. Keboola says the older kbc CLI remains supported and that new command-line work should use kbagent. Connector catalogs were not opened. https://www.keboola.com/llms.txt is a product index for agents. Computer-use viability is unassessed. MCP and CLI both score 8. MCP is the best path because the hosted server does not require the local kbagent binary. The Storage API scores 7.

Best path today: mcp · Overall automatability: 8/10