How to automate video editing with AI (2026 guide for editors)
You can automate video editing with AI by handing repetitive tasks to tools: transcription and text-based cuts (Descript, Premiere), footage logging and search (Jumper), and multi-step jobs to AI agents like Claude or Codex, which connect to Premiere Pro, DaVinci Resolve, Final Cut Pro or Avid through MCP to pull selects and build sequences. You still make the creative calls.
AI can already take over a lot of the repetitive work in editing: transcription, logging, finding shots, and first assemblies. It can’t make the creative calls. Here’s what each kind of tool does, and how to connect an AI agent to your NLE.
What AI can automate today
| Task | Example tools | What you still check |
|---|---|---|
| Transcription | Premiere, Resolve, Descript, Jumper | Names, jargon, and who’s speaking |
| Logging and tagging | Jumper, MAMs like Iconik or Axle | That the description matches the shot |
| Finding shots and B-roll | Premiere Media Intelligence, Resolve IntelliSearch, Final Cut, Jumper | Taste, continuity, and rights |
| Finding quotes | Descript, NLE transcripts, Jumper speech search | That the line means what you think |
| Rough cut from a transcript | Descript, Premiere text-based editing, Jumper Script Editor | Pacing and story |
| Assembly from a written brief | An AI agent through Jumper’s MCP server | Everything. It’s a first assembly. |
| Captions and reframing | Built-in NLE features, Descript | Line breaks and framing |
Three levels of automation
- Single-task features in your NLE. Premiere, Resolve Studio, and Final Cut can transcribe, search, and reframe. Each feature does one job in the open project.
- AI-assisted editing. You search, the tool returns moments, and you decide what stays. Jumper’s speech search and Script Editor work this way, and so does Descript.
- Agentic editing. You write a brief, and an agent like Claude or Codex runs the searches and builds a sequence for you to review. More in the AI agents guide.
How agents connect: MCP
MCP (Model Context Protocol) lets an AI app call tools. Jumper runs a local MCP server, so an agent can search your analyzed footage, read the open sequence, and send results back to your NLE. Your footage isn’t uploaded. A cloud model like Claude receives file paths, timecodes, transcript excerpts, search results, and downsampled thumbnails, but not the video files. To keep all of it on your machine, use a local model through LM Studio.
Some NLEs also have MCP servers that control the app itself. Resolve Studio 21.1 ships one, and there are community servers for Premiere. Jumper’s server covers the footage side, and you can use both.
Connect Claude or Codex to your NLE
- Open Jumper, go to Settings, and find AI Agents.
- Install the connection for Claude Desktop, Claude Code, or Codex Desktop, or copy the MCP URL into another agent.
- Ask the agent whether it can use Jumper and whether your media is analyzed.
- Give it a brief and a destination, then check the result.
What the agent can do depends on the NLE:
- Premiere Pro MCP: reads the open sequence and builds new sequences in Premiere.
- DaVinci Resolve MCP: builds timelines in Resolve and works alongside Resolve’s own MCP server.
- Final Cut Pro MCP: exports FCPXML projects for you to import. It can’t edit the open timeline yet.
- Avid Media Composer MCP: reads the open sequence and writes edits back as AAF.
Example briefs
- “Find interview answers about the harbor closure and matching wide shots of the port, and put them on a new timeline.”
- “Pull every hero shot of the device on a white background from the 2025 product shoots into a selects sequence.”
- “Find scenes where these two cast members are together and talking about the vote.”
- “Find the strongest answers about the tour, with the guest on camera, and make a selects sequence.”
An agent can handle a theme like “the harbor closure” by running searches and reading the transcripts and summaries Jumper creates. It’s good at finding and assembling, and bad at story, pacing, and taste. Treat every sequence as a first assembly, and check the source before anything airs.
Descript or an agent?
Use Descript for talking-head and podcast pieces you’ll finish in Descript. Use an agent with Jumper when you cut in Premiere, Resolve, Final Cut, or Avid and the footage is a large local library. Plenty of jobs use both.
If you run the same brief every week, you can script it against Jumper’s local API. See vibe coding for video editors.
FAQ
Can Claude or ChatGPT edit my Premiere Pro timeline?
Yes, through MCP. Jumper's MCP server lets Claude search your analyzed footage and send sequences back to Premiere. ChatGPT can do the same if your client supports MCP. A chat with no tool connection can't see your timeline or your drives.
How do I connect an AI agent to DaVinci Resolve?
In Jumper, open Settings, go to AI Agents, and install Claude or Codex, or paste the MCP URL into another agent. Resolve Studio also has its own MCP server for controlling the app, and you can use both.
Can an AI assistant control Final Cut Pro?
Through Jumper, an agent can search your analyzed footage and export an FCPXML project for you to import. It can't edit the open Final Cut timeline yet.
Is my footage uploaded to the AI company?
No. Jumper analyzes footage locally. A cloud model receives file paths, timecodes, transcript excerpts, search results, and downsampled thumbnails, but not the video files. A local model, for example in LM Studio, keeps all of that on your machine.
Do I need to code?
No. Install Claude or Codex from Jumper's AI Agents settings and write your brief in plain English. Coding against Jumper's local API is optional, for automations you want to reuse.
Will AI replace editors?
No. It shrinks logging, hunting for shots, and building first string-outs. Story, pacing, taste, and responsibility for what airs stay with the editor.
Try Jumper free and run the workflow on a project you already know.