Claude highlight edit from a long video: Resolve 21.1 vs Sume MCP

Blackmagic says DaVinci Resolve 21.1 supports AI assistants such as Claude, Claude Code and ChatGPT Codex, which can create highlight edits from long form video inside Resolve. Sume is a separate, cloud-side route: an assistant connected to Sume's hosted MCP can inspect a clip, trim ranges and join them, without Resolve or a project file.

Tool names are from MCP tools and gates, read 2026-10-01. Sume does not choose the highlights; your assistant picks the timestamps.

What are the steps in Sume?

Import the long file to media.sume.com (media-imports_create). Call video_inspect for probe facts and stills, optionally with transcribe: true so the assistant can read what is said. Pick ranges, call video_trim for each, wait with jobs_wait, then pass the new files into timeline_create video[] with source_in 0. The timeline needs an audio spine: set audio.duration_seconds and either an audio url or audio.mode: "silence". Writes need an idempotency_key.

How many trim jobs will a reel need?

One per range. script_run can loop calls of the same shape in one turn, within its documented max_calls and max_paid_calls bounds, and the response lists the child jobs[] to wait on. Read ids and results with jobs_wait and jobs_result, as in waiting for long video jobs.

What does Sume not do here?

It does not read or write Resolve projects, timelines or media pools, and the docs describe no highlight scoring tool on production: video_analyze and video_segment appear only on the dev host when enabled. Semantic choices come from the assistant using stills and transcript. For plain splitting by time see splitting a long video into clips.

Where does the work run?

Trim and timeline render run as Sume jobs on the worker media runtime, so closing your chat does not stop a submitted job; read it later by id. Your machine only needs to reach mcp.sume.com.