Kestra 2.0 overview: MCP tools and upgrade requirements

This release overview explains Kestra 2.0's architecture and workflow features. The presenter says workers now communicate with a controller over gRPC rather than requiring direct database access. Separating the queue from the database allows different backend choices, while workers can deploy across regions and clouds.

The MCP server lets an AI agent call Kestra flows as tools through an MCP tool trigger and receive data after execution. Claude Code is the example client. AI Copilot has a persistent sidebar for questions about executions and logs, with editing, questioning and planning modes and access to previous conversations.

The governance features include more granular RBAC permissions and policies that check compliance or block flows at namespace and tenant levels. Cases track incidents alongside failed executions, and Promote synchronizes flows between environments. Worker groups route tasks by tags. The Loop task uses isolated sub-executions, which the presenter says improves stability when processing large datasets.

A slim Docker image installs essential plugins first and adds others when used. Plugin artifacts let plugins load interface components at runtime. For upgrades, the speaker specifies version 1.3 as a prerequisite for 2.0 and recommends the separate migration guide for required flow changes. The release is described as LTS with one year of support.