FFmpeg you can talk to.
The open-source, local-first AI video editor — chat, CLI, or browser timeline.
Drag five screen recordings into a browser tab, type "trim each to the most interesting 10 seconds, add title cards between them, render to 1080p" — and Claude calls 19 deterministic FFmpeg tools to do it. No cloud, no account, no telemetry, no generative AI rewriting your audio. Just FFmpeg, named tools, and an op log you can
git diff.
As a Claude Code skill (chat-driven, no global install):
npx skills add MakeMyClip/editorAs a CLI (for scripts and CI):
npm i -g @makemyclip/editor
clip --helpAs a browser UI (visual timeline + chat panel):
clip ui # opens http://127.0.0.1:5573The Claude Code skill auto-discovers triggers and shells out via npx -y on demand. The chat panel inside clip ui needs ANTHROPIC_API_KEY; everything else works offline.
Five screen recordings → one polished product demo. Type this into Claude Code or the browser chat panel:
Import these five recordings. Trim each to the most interesting 8–12 seconds. Add black title cards saying "Demo 1: Inbox", "Demo 2: Search", "Demo 3: Compose", "Demo 4: Calendar", "Demo 5: Reports" between them. Concat in order. Render to 1080p mp4.
The agent runs roughly ingest × 5 → trim × 5 → add_title_card × 5 → concat → render. The final mp4 lands in your workspace folder, every step is appended to session.json, and you can ⌘Z any of it.
Or skip the chat and do it from a script:
clip ingest demo1.mp4
clip trim demo1.mp4 5 20
clip add_title_card demo1-trim.mp4 "Demo 1: Inbox"
clip concat demo1-with-card.mp4 demo2.mp4 demo3.mp4
clip render concat-output.mp4 mp4 23 fast 1920Same registry, same session log, same output.
19 tools across 5 categories, three surfaces (Claude Code skill, CLI, browser UI), one shared op log.
| Category | Tools |
|---|---|
| Cut & arrange | trim, split, concat, transition (12 kinds) |
| Text & captions | add_text, add_title_card, add_captions |
| Visual primitives | transform, adjust, zoom_pan (Ken Burns), overlay, speed |
| Composites & specialty | highlight_reel, silence_remove, chroma_key, stabilize |
| Output & I/O | ingest, preview, render, add_audio |
Plus four meta-ops not in the registry: snapshot, undo, inspect, delete.
The browser UI adds drag-drop import, a horizontal timeline, a render queue, a chat sidebar, snapshot/undo buttons, and keyboard shortcuts (⌘Z undo, ⌘S snapshot, ⌘⇧N new op, Esc cancel).
Existing tools force a tradeoff that doesn't need to exist:
- iMovie / CapCut / DaVinci Resolve — mouse-and-timeline desktops. Can't be driven by an agent, scripted in CI, or version-controlled.
- Descript — closed-source, cloud-based, rewrites audio with generative AI you can't inspect.
- Runway / Pika / Luma — generate video from text. Useful for VFX, useless for assembling a 90-second demo from recordings you already have.
- Raw FFmpeg — deterministic and free, but the surface is hostile to LLMs: cryptic filter graphs, no session model, no undo, no tool catalog.
MakeMyClip Editor closes the gap: deterministic FFmpeg editing with a 19-tool catalog any agent can call, a session log it can inspect, and a browser UI a human can drive when the agent goes off-script.
Full comparison matrix — features and use-cases vs. iMovie, Descript, Runway, raw FFmpeg
| Feature | MakeMyClip Editor | iMovie / CapCut | Descript | Runway / Pika | Raw FFmpeg |
|---|---|---|---|---|---|
| Agent-driven (chat to edit) | ✅ | ❌ | ❌ | ❌ | |
| Open-source (MIT) | ✅ | ❌ | ❌ | ❌ | LGPL/GPL |
| Local-only (no cloud) | ✅ | ✅ | ❌ | ❌ | ✅ |
| Deterministic output (no AI generation) | ✅ | ✅ | ✅ | ❌ | ✅ |
| Free forever | ✅ | ✅ | freemium | paid | ✅ |
| Programmable CLI | ✅ | ❌ | ❌ | API | ✅ |
| Visual timeline UI | ✅ | ✅ | ✅ | ✅ | ❌ |
| Snapshot & undo | ✅ | ✅ | ✅ | ❌ | |
| Inspectable session log (JSON) | ✅ | ❌ | ❌ | ❌ | ❌ |
| Works as a Claude Code skill | ✅ | ❌ | ❌ | ❌ | manual |
| Stream-copy for lossless cuts | ✅ | ❌ | ✅ | ||
| Zero telemetry | ✅ | ❌ | ❌ | ❌ | ✅ |
| Job to be done | MakeMyClip Editor | iMovie / CapCut | Descript | Runway / Pika | Raw FFmpeg |
|---|---|---|---|---|---|
| Trim screen recordings to highlights | ✅ | ✅ | ✅ | ❌ | ✅ |
| Assemble a product demo from N clips + title cards | ✅ | ✅ | ❌ | ||
| Add script-provided timed captions | ✅ | ✅ | ❌ | ||
| Auto-transcribe & edit text-as-video | ✅ | ❌ | ❌ | ||
| Auto-cut silence from a recording | ✅ | ❌ | ✅ | ❌ | |
| Chroma-key / green-screen compositing | ✅ | ✅ | |||
| Generate b-roll / VFX shots from a prompt | ❌ | ❌ | ✅ | ❌ | |
| Multi-track audio mixing with effects | ❌ | ✅ | ❌ | ||
| Drive editing from an AI agent in natural language | ✅ | ❌ | ❌ | ||
| Run as a scriptable CI pipeline | ✅ | ❌ | ❌ | ✅ | |
| Long-form (>30 min) with hundreds of clips | ✅ | ✅ | ❌ | ✅ | |
| Edit on Linux / headless / fully offline | ✅ | ❌ | ❌ | ❌ | ✅ |
Legend: ✅ great fit ·
Honest summary: MakeMyClip Editor is the right pick when you want deterministic, scriptable editing that an AI agent can drive end-to-end. Not the right pick for generative video (Runway), text-as-video transcript editing (Descript), or a heavy multi-track NLE (DaVinci, Premiere).
Architecture — three surfaces, one registry, one append-only op log
Claude Code → skill triggers → npx -y @makemyclip/editor <tool>
│
Browser UI → /api/tools/:name → TOOL_REGISTRY (19 tools)
│
Chat panel → /api/chat → AI SDK + Anthropic + tool dispatch
│
▼
Tool handlers (TypeScript)
│
▼
FFmpeg subprocess (args as array, no shell)
│
▼
session.json (append-only op log)
The session is the source of truth: every editing operation appends one entry with { id, tool, args, result, timestamp }. The CLI, browser UI, and chat panel all write through the same appendOp() path, so any combination of human + agent edits stays consistent.
- Language: TypeScript (Node 24+) and React 19 (browser UI)
- Timeline schema: Zod (shared with the MakeMyClip.com web app)
- Subprocess: execa — args as an array, no shell injection
- FFmpeg: bundled via
ffmpeg-static, withMAKEMYCLIP_FFMPEG_PATHoverride or system-binary fallback - AI SDK: Vercel AI SDK +
@ai-sdk/anthropic(chat panel only) - UI: Hono server + Vite + React, plain CSS
FAQ — setup, capabilities, privacy, AI integration, comparisons, license
Do I need FFmpeg installed separately?
No. MakeMyClip Editor bundles FFmpeg via ffmpeg-static, so installing the npm package or running npx skills add MakeMyClip/editor is all you need. For your own FFmpeg build (LGPL-only, custom codecs, hardware acceleration), set MAKEMYCLIP_FFMPEG_PATH to its path and the bundled binary is ignored.
What platforms does it run on?
macOS, Linux, and Windows — anywhere Node 24+ runs. The browser UI binds to 127.0.0.1 and needs no internet connection.
Can I use it without Claude Code?
Yes. The clip CLI and clip ui browser app are first-class surfaces. Claude Code is one of three entry points — the editor is agnostic to which agent (or human) drives it.
Does it require an API key?
Only the chat panel inside clip ui needs ANTHROPIC_API_KEY. The CLI, the browser UI's visual editing, the Claude Code skill, and every editing tool work without any API key.
Is it production-ready? This release is feature-complete for local editing — 19 tools, 356 tests passing, browser UI shipped, chat panel shipped. The API surface is still pre-1.0 (tool schemas may change in minor ways before 1.0). Use it for real work; pin a version in CI.
What's the maximum video size or duration? No hardcoded limit. FFmpeg streams the file rather than loading it into memory, and the browser UI streams uploads to disk — multi-GB recordings work fine. This release isn't tuned for projects with hundreds of clips or runtime above 30 minutes; for those, drive the CLI directly.
Does it support 4K, HDR, or vertical (TikTok / Reels) formats?
4K and vertical aspect ratios work out of the box — render takes a maxWidth and preserves the source aspect ratio. HDR pass-through works; deliberate HDR-to-SDR tone-mapping isn't a named tool yet.
Does it support MCP (Model Context Protocol)?
Not yet. The 19-tool registry is already MCP-shaped ({ name, schema, fn }), so adding the MCP transport is small when the demand surfaces — tracked for 1.0.
Does it send my video files anywhere? No. The editor makes zero network calls in its core. The optional chat panel sends conversation text (not video files) to Anthropic's API when enabled.
Where does it store my files?
In a local workspace folder — $TMPDIR/makemyclip-editor by default; override with MAKEMYCLIP_WORKSPACE. The folder holds input files, output files, session.json, chat.json, and snapshots/. You can version-control the whole folder.
Is my chat history private?
Chat history is stored locally in workspace/chat.json — nothing leaves your machine except what you explicitly send. Each chat turn forwards your messages plus a summary of recent session ops to Anthropic's API; responses stream back.
Can I use a different AI model (Gemini, OpenAI, local LLMs)?
The chat panel uses claude-sonnet-4-6 via the Vercel AI SDK, which supports Gemini, OpenAI, Mistral, local Ollama, and others through provider imports. Swapping is one import + one config change in src/ui/server.ts; a UI-level model picker is on the roadmap. The CLI is model-agnostic.
How does the agent know which tool to call?
Each of the 19 tools ships with a Zod input schema (parameter descriptions inline via .describe()) and a one-line tool description. The chat panel passes the full catalog to the model via the AI SDK's tool() interface; the model reads the descriptions, picks a tool, and fills in the arguments.
What if the agent makes a mistake?
Every edit appends to session.json and produces a new output file in the workspace; nothing is destructive. Press ⌘Z to pop the last op; click Snapshot before risky operations to save a named restore point; the entire workspace is a folder you can git init.
Is it free?
Yes — MIT-licensed, free forever for local editing. No paid tier, no signup, no upsell in the editor. Paid AI-generation features (voice synthesis, music, stock footage) will live on MakeMyClip.com when they ship, behind a separate @makemyclip/generation package.
What's the license situation with the bundled FFmpeg?
The MakeMyClip Editor source code is MIT. The bundled FFmpeg binary (via ffmpeg-static) is GPL-licensed because it includes codecs like libx264 and libx265. Subprocess invocation keeps GPL terms confined to the binary, not your application code. For LGPL-only or custom-build requirements (e.g. iOS App Store), set MAKEMYCLIP_FFMPEG_PATH to your own binary.
How does it compare to Descript? Descript edits text-as-video using generative AI to rewrite audio. MakeMyClip Editor edits video as video, using deterministic FFmpeg operations — every cut is exact, lossless where possible, reproducible. Descript is closed-source and cloud-only; MakeMyClip Editor is MIT and local.
How is this different from running FFmpeg directly? FFmpeg is one binary with a complex CLI; MakeMyClip Editor is 19 named tools with typed inputs, a session log, snapshot/undo, a browser UI, and an agent-callable API. You're still running FFmpeg under the hood — but with a catalog any LLM can call, input validation that prevents bad commands, and a log you can inspect or version-control.
What's the relationship between this editor and the MakeMyClip.com website? The editor (this repo) is MIT-licensed, free, local, and limited to deterministic FFmpeg operations. MakeMyClip.com is the separate hosted product where paid AI generation features will live. The two share the timeline schema, so projects move between them.
How do I add my own editing tool?
Add a file in src/tools/<name>.ts exporting a Zod input schema and a handler, register it in src/ui/tool-registry.ts, and the tool becomes available in the CLI, the browser UI's tool picker, and the chat panel's agent. See AGENTS.md for conventions.
✅ Great fit for: assembling product demos, screen recordings, tutorials, and social clips with an AI agent doing the labor · scripting repeatable video pipelines in CI · sharing version-controllable timeline JSON · editing offline.
❌ Not the right tool for: frame-accurate color grading (use DaVinci Resolve) · multi-track audio mixing with effects (use Reaper, Logic) · generative-AI video (use Runway, Pika) · long-form projects with hundreds of clips.
- Now — feature-complete local editing: 19 tools, browser UI, chat panel, snapshot/undo
- Next — model picker, SSE for live session updates, AI Elements / shadcn migration
- 1.0 — frozen tool schemas, MCP transport, published Anthropic skill, docs site
- Beyond — desktop app (Electron), cloud rendering via MakeMyClip.com
The full milestone log lives in PR descriptions on GitHub.
Built on the work of FFmpeg, the Vercel AI SDK, Anthropic Claude, Zod, Hono, Vite, and React.
| npm | @makemyclip/editor |
| Skill page | skills.sh/MakeMyClip/editor |
| GitHub | MakeMyClip/editor |
| Website | makemyclip.com |
| For AI agents | llms.txt |
| Issues | GitHub Issues |
MIT for the editor source; GPL for the bundled FFmpeg binary (subprocess-isolated). Use it, fork it, ship it.