Claude Code video editing

Connect Claude Code to a real video editor in about five minutes, then hand it whole jobs: a long-form edit, a batch of Shorts, a narrated screen recording. No coding required.

Fastest path: open PandaStudio and go to Settings → Local automation → Connect agents for one-click setup. The steps below are the manual route if you prefer to wire it up yourself.

Why Claude Code?

Claude Code is Anthropic's official terminal agent. PandaStudio has an official Skill for it — a set of instructions that teaches Claude the full command surface so it can transcribe, caption, add motion graphics, and export without you having to explain the API. This is the recommended setup.

Connecting Claude Code, start to finish.

Install PandaStudio

Download PandaStudio from writepanda.ai and run the installer for your OS (macOS or Windows). Open it once so it finishes setting up the recordings folder.

Install Node.js (if you don't already have it)

Claude Code and the PandaStudio CLI both need Node 18+. Check:

bash

node --version

If that prints v18.x.x or higher you're set. Otherwise grab the LTS installer from nodejs.org.

Install Claude Code

bash

npm install -g @anthropic-ai/claude-code

Then run claude once — it walks you through signing in with your Anthropic account.

Install the CLI + Skill from PandaStudio Settings

Open PandaStudio, go to Settings → Local automation, and click both buttons:

Install (under Shell command) — makes the pandastudio command available in any terminal (the binary ships inside the app — no separate npm install needed).
Install Skill — copies the Skill into Claude Code, Codex, OpenClaw, and Hermes in one click. For Claude Code it lands at ~/.claude/skills/pandastudio/SKILL.md and Claude picks it up automatically — no restart needed.

Prefer the terminal? Same two installs are reachable as npm install -g @writepanda/cli and npx skills add kamskans/pandastudio-skills (the latter also covers Cursor, Cline, Continue, and 40+ other agents).

Turn on Local automation, then click Connect

Open PandaStudio.
Go to Settings → Automation and check that Local automation is on.
Scroll to Connect agents, find Claude Code, and click Connect. PandaStudio registers its MCP server with Claude Code for you.
If Claude Code was already open, start a new session — it reads its MCP servers at startup. Type /mcp to see PandaStudio listed as connected.

The server binds to 127.0.0.1 only and requires a bearer token that PandaStudio writes locally. Nothing outside your machine can reach it.

Try your first prompt

Open any terminal, run claude, and try:

"Open my latest recording, transcribe it, remove filler words and silences, then export an MP4."
"Create a new 9:16 project, import the file at ~/Recordings/launch.mov, add bold captions, and preview it."
"List all my projects from this week and tell me which one is longest."

Claude will call pandastudio system.status to confirm the license, then chain commands like transcript.transcribe, transcript.remove-fillers, caption.set-template, and export.start — narrating each step as it goes.

What to ask Claude Code once it's connected

Four jobs to start with. Each one runs through PandaStudio's commands, so every step lands on a normal timeline you can open, preview and adjust by hand afterwards.

Edit a long talking-head video

Claude transcribes, removes fillers and silences, applies an edit style (captions, zooms, chapter cards) and exports.

"Open my latest recording, clean it up, and edit it in the Ali style recipe. Then export a 1080p MP4 and write a title, description and chapters."

Turn one long video into Shorts

Claude reads the transcript, picks the strongest moments, makes a 9:16 project for each with the face kept in frame, and styles them.

"Find the three best moments in this episode for YouTube Shorts. Make each a vertical Short in the storytime style with captions, and export them."

How clipping works

Record your screen with an AI voiceover

Claude starts a PandaStudio screen recording, walks through your app (it needs a browser tool such as Playwright or Claude in Chrome for the clicking), stops it, then writes and adds a narration with the on-device voice.

"Record my screen while you open localhost:3000 and walk through the signup flow. Then write a short voiceover for it, add it, cut the dead air and export."

Agent screen recording in detail

Publish to YouTube

Claude generates the thumbnail, title and description, then uploads to your connected channel.

"Make a thumbnail for this video, then publish it to my YouTube channel as unlisted so I can check it."

Why use an editor instead of letting Claude script ffmpeg?

Claude Code can edit video on its own by writing ffmpeg commands, and open-source projects do exactly that. It works for simple cuts. The trouble starts when you want to see the result before it renders, fix one caption, or move a zoom by half a second: with a script, every change is another round of prompts and a full re-render.

With PandaStudio, Claude's edits become regions on a timeline. You watch the preview, drag the one thing that's off, and export. The heavy lifting runs on your machine: transcription, filler removal, background removal, audio clean-up and the narration voice are all local models, so Claude spends its tokens on decisions, not on processing video.

Troubleshooting

"License gate engaged" — open PandaStudio and start a free trial export, or paste in your license key from the pricing page.
"Connection refused" — the automation server toggle in Settings is off, or PandaStudio isn't running.
Skill not loading — run npx skills add kamskans/pandastudio-skills again. The tool re-copies to the correct path idempotently.

Reference

CLI & MCP commands

Browse all 60+ verb-noun commands the agent can call.