Kling is now piloted from AI agents, in MCP or CLI

Kling AI opens direct model access to AI assistants via a new MCP server and CLI tool, bypassing its web interface for paid-credit image and video tasks.

Kling AI is opening access to its image and video generation models from AI assistants, bypassing its web interface. The company is releasing two integration components: a Model Context Protocol (MCP) server and a command-line interface (CLI) tool.

The MCP connector plugs into compatible assistants like Cursor or Claude. Once authorized, the user describes their request in natural language ("turn this image into a five-second video with cinematic camera movement") and retrieves the output directly within the assistant. The CLI exposes the same capabilities as terminal commands, usable in scripts, continuous integration (CI) pipelines, ComfyUI nodes, or simple shell sequences. Both require the installation of a Kling skill, with the documentation recommending only one be used: with both in place, the assistant risks drawing from one or the other unpredictably.

Several conditions govern usage. MCP and CLI only work with paid credits from the personal workspace, excluding free off-peak generations from official platforms, and without the benefits of the team workspace for now. The rate is capped at 5 QPS, with only one simultaneous video task for non-subscribers. Finally, generated files remain accessible for twenty-four hours via their URL, and a task cannot be canceled once launched.