Skip to content

MakeMyClip/editor

Repository files navigation

MakeMyClip Editor

MakeMyClip Editor

AI-native video editor — talk to Claude to make video.

Trim, zoom, caption, and assemble clips by asking Claude. Ships as a Claude Code skill + a clip CLI for direct use from terminal, scripts, or any agent that can shell out.

Status: pre-alpha. Nine tools shipping (ingest, trim, split, concat, transition, add_text, add_audio, preview, render); more landing iteratively. Star to follow along.


What it is

An open-source video editor you drive by chatting with an AI agent. No timeline UI to learn — describe the edit in plain English and the agent calls the editor's tools to do it.

You:    "Trim the first 12 seconds, zoom on the cursor at 0:34, and add a caption."
Agent:  *calls trim, zoom_pan, add_text → renders → returns clip.mp4*

Under the hood it's a thin, safe wrapper around FFmpeg with a structured timeline schema. The agent does the planning; the editor does the rendering.

Why

Existing video editors are mouse-and-timeline. Existing AI video tools are black-box generators. Neither fits the workflow of a developer or founder who wants to:

  • turn a 5-minute screen recording into a 45-second product demo,
  • caption and zoom a tutorial without opening Premiere,
  • assemble a social clip from raw footage in one prompt.

This editor is the missing piece: deterministic, local, scriptable editing that any AI agent can drive. Every edit is a structured timeline document the agent can inspect, version, and hand off — not an opaque sequence of FFmpeg commands.

Install

Claude Code — one command:

npx skills add MakeMyClip/editor

No global install, no config edits, no client restart. The skill auto-discovers triggers and shells out to the CLI on demand (npx -y downloads the package on first use, cached after).

Terminal, scripts, CI:

npm i -g @makemyclip/editor

Then use the clip CLI directly. Run clip --help for usage.

MCP server for other clients (Cursor, Claude Desktop, Continue) is on the roadmap — see the project shape for what's currently in vs out.

When to use it

✅ Use it when you want to:

  • Edit screen recordings, demos, tutorials, or social clips with an AI agent
  • Script repeatable video pipelines (one prompt = one clip)
  • Share an editable timeline JSON with your team

❌ Don't use it (yet) for:

  • Frame-accurate color grading or VFX work
  • Multi-track audio mixing
  • Long-form (>30 min) projects

Capabilities (v1)

Tool What it does
trim Cut clips by timecode
concat Stitch clips together
zoom_pan Ken Burns / focus zoom on a region
add_text Captions, titles, lower-thirds
add_audio Background music, voiceover overlay
transition Crossfade, cut, dip-to-black
render Export to MP4 / MOV / WebM
preview Generate an HTML scrubbable preview

All edits are non-destructive — the agent builds a timeline JSON, you can inspect and tweak it, then render.

Architecture

Claude Code  →  skill auto-discovery  →  npx -y @makemyclip/editor <tool>
                                              │
                                              ▼
                                         clip CLI (tool handlers)
                                              │
                                              ▼
                                    FFmpeg subprocess (stream-copy / filter)

The skill is intent-matching markdown; the CLI is the single execution surface. No persistent server, no daemon, no client-specific wiring — just one process per invocation.

  • Language: TypeScript (Node 24+)
  • Timeline schema: Zod (shareable with the MakeMyClip.com web app)
  • Subprocess: execa — args as an array, no shell injection
  • FFmpeg: bundled via ffmpeg-static, with $MAKEMYCLIP_FFMPEG_PATH override or system-binary fallback

Free & open source

The editor is MIT licensed and free forever for local editing. Anything FFmpeg can do, this does for free — no account, no telemetry, no limits.

Paid AI-generation features (voice, music, stock, premium templates) will live on MakeMyClip.com when ready.

License

The MakeMyClip Editor source code is MIT licensed — use it, fork it, ship it.

The bundled FFmpeg binary (via ffmpeg-static) is GPL licensed because it includes codecs like libx264 and libx265. This is fine for personal use, open-source projects, internal company use, server-side SaaS, and most commercial desktop products — your own code stays MIT, and the subprocess invocation pattern keeps the GPL terms confined to the FFmpeg binary itself, not your application code.

If your situation requires an LGPL-only or custom FFmpeg build (e.g. strict no-copyleft enterprise policy, iOS App Store distribution), set MAKEMYCLIP_FFMPEG_PATH to your own binary and the bundled one will be ignored.

Links

About

AI-native video editor — talk to make video. Trim, zoom, caption, and assemble clips from any AI agent (Claude, Cursor, Copilot). Delivered as an MCP server + agent skill.

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors