Claude or ChatGPT, Resolve and Prewatch

1. Get Prewatch from the Mac App Store. It runs on Macs with Apple silicon and macOS 26 or later.
2. Start the free trial. 14-day free trial, then $59 a year or $7.99 a month.
3. Open Prewatch.

1. Choose File → Add Folder to Library… (⌘O) and pick a folder of footage: on the Mac, an external drive or a NAS. You can add several.
2. Turn on Watch and Index for the folder, in the folder view or by right-clicking it in the sidebar. Until you do, Prewatch only reads the folder.
3. Leave Prewatch open; it can be hidden. It indexes every clip and keeps up as new files arrive. A file that is still being copied is indexed once it stops changing.

Where the index goes. By default into the Prewatch folder inside your footage folder, so the index travels with the media to another drive, a NAS or another Mac. To keep it on this Mac instead, and use it while the drive is unplugged, choose Settings → Storage → Library on this Mac.

Speech. The first clip with speech downloads the Whisper speech model once (Large v3 Turbo, 1.6 GB, from huggingface.co). To fetch it now: Settings → Indexing → Download Now.

Prewatch adds only its own entry and keeps the other servers in those settings. You never edit a configuration file by hand.

ChatGPT in a web browser can't reach apps on your Mac; the ChatGPT Mac app can. For Claude Code or the Codex CLI, use Copy Command on the same page and paste it into Terminal.

1. Update to DaVinci Resolve Studio 21.1 or later. Trade press reports its MCP server as a Studio feature.
2. Open Resolve's setup for AI assistants (reported under File → Setup AI Assistants; the Resolve 21.1 reference manual has a section on it) and add Resolve to Claude Desktop or to Codex in the ChatGPT app. Reports say the setup detects both.
3. Quit and reopen Claude Desktop or the ChatGPT app.
4. Keep Resolve open, with the project you want to cut in.

Prewatch never talks to Resolve. Your assistant uses both: Prewatch to find the material, Resolve's server to build the timeline. How the two fit together, with sources.

1. Turn on Describe clips with a vision model.
2. Choose the Provider, Claude (Anthropic) or OpenAI (ChatGPT), and a Model.
3. Create an API key in the Anthropic Console or on platform.openai.com. A Claude or ChatGPT subscription does not include API use; Anthropic or OpenAI bills it to your account.
4. Paste the key, click Save to Keychain, then Test key. The test spends no tokens.

What it adds. A written description of each clip, moments with in and out points, shot roles and better translations. Without a key, transcripts, speakers, camera motion, quality flags, tags, days, scenes and search all work.

What it costs. The page shows what a typical clip costs with the model you chose. Once a key is saved, new clips are described as they are indexed, and clips already indexed in watched folders are described in the background, at most 200 per run (change it under Existing clips). Describe existing clips… shows the clip count, requests and estimated cost before you confirm.

What is sent. Each clip's contact sheet and keyframes as still images, and the text of a transcript in another language for its translation. Never video, audio, locations or speaker labels. The key stays in your Keychain and goes only to Anthropic or OpenAI.

1. Start a new chat. prewatch should be listed among the chat's tools or connectors, with Resolve beside it.
2. Ask about your footage: “What happened on day 3?”, “Find the clips where we talk about the route.” or “Which clips are flagged as black or frozen?”
3. When the app asks to use a Prewatch tool, allow it once or always.

1. Ask for an edit: “Make a string-out of day 1 in capture order” or “Cut a 60-second teaser from the best moments.”
2. The assistant picks clips and in and out points with Prewatch, then imports and places them through Resolve's MCP server.
3. Review the timeline in Resolve and keep refining in the same chat: “Swap the second shot for a wider one.”