Skip to main content
With the Kino MCP server connected, an agent can take an edit from “find the good takes” to a rendered file. This guide walks through what that loop looks like in practice.

​
Start from a prompt

Open your agent with Kino connected and describe the edit:
Make a 60-second cut of the interview where Sarah talks about the product launch. Use the b-roll from last week for cutaways.
The agent reads the Kino workflow guide first (getKinoInstructions), then either edits the open project or opens/creates one (listProjects / openProject / createProject). Project tools error if nothing is selected. Ask for a format — “a vertical version for TikTok” — and the agent either creates the project at that frame size or reformats the open one with setCompositionSettings.

​
The agent finds your footage

Instead of you scrubbing through clips, the agent searches:
  • transcriptSemanticSearch finds where Sarah talks about the launch, with timestamps
  • visualSemanticSearch finds b-roll by what’s on screen
  • readVideo pulls the full transcript and highlights of a clip so the agent can pick exact in and out points
If you already selected clips in the Kino UI, the agent can work from your timeline and media-library selections instead (getSelectedItems).

​
The agent edits the composition

For timeline surgery — moving clips, trimming cuts, splitting, ripple edits — the agent uses structured operations: timelineGet to read the timeline and timelineEdit to change it, with the same guarantees as editing by hand. Everything else is code. Kino compositions are React, so the agent edits them the way it edits any codebase: reading files, searching them, and applying patches (projectRead, projectGrep, projectPatch). Inserting a title is a code change. So is anything Kino doesn’t have a button for yet: if the agent can compute it (a beat grid, a custom color effect), it can write it into the timeline. After each change it can render a screenshot of the timeline at a specific timestamp (projectScreenshot) to verify the result visually, and profile playback (projectProfile) if it added heavy effects. Your timeline updates live in the desktop app while this happens. You can watch the cut come together and interrupt with notes at any point.

​
Export

When you’re happy with the cut:
Export this as launch-cut-v1.mp4 on my Desktop
The agent starts the render (exportVideo) and polls until the file is ready (getVideoExportTask).

​
Tips

  • Be specific about duration, tone, and which footage to use. Vague requests produce vague searches.
  • Ask for a screenshot whenever you want to sanity-check the state of the timeline without switching windows.
  • The agent works on whatever project is open in the desktop app, so keep the right project open when you start.