OpenCut Rewrites Its Open-Source Video Editor from Scratch, Making AI Agents First-Class Citizens

A Counterintuitive Decision

In 2025, OpenCut was already the hottest open-source video editor on GitHub—a free CapCut alternative with hundreds of thousands of users and daily traffic on opencut.app. Then the team made a counterintuitive call: rewrite it from scratch.

Not incremental refactoring—a full rebuild. The old version was renamed opencut-classic, and the new version is being written from zero. opencut.app still runs the old version while the new one grows slowly at new.opencut.app.

Rewrites are famously the thing you're "never supposed to do" (per Joel Spolsky's classic 2000 essay). But the team is betting that video editors need to be re-architected for the AI era, and the old architecture can't be evolved to get there.

Six Promises of the New Architecture

The rewrite's README lists six goals:

1. Editor API — every editor capability exposed via API 2. Plugin-first architecture — third-party plugins as first-class citizens 3. Unified cross-platform codebase — desktop, mobile, and browser sharing one Rust core 4. MCP server — an interface for AI agents 5. Headless mode — automation and batch rendering 6. In-editor scripting

Together these point in one direction: the video editor evolves from "a GUI tool for humans" into "a programmable rendering platform."

MCP Server: Letting AI Agents Edit Video

This is the most interesting item. MCP (Model Context Protocol), proposed by Anthropic in 2024, lets AI agents call external tools. A built-in MCP server means agents like Claude Code and Codex can operate the video editor directly:

• "Cut this 10-minute video into five 2-minute shorts" — the agent calls the cut API
• "Add subtitles using Whisper transcription" — transcription + subtitle APIs
• "Batch-add intros and outros to 100 videos" — headless batch rendering
• "Swap the background music for something more upbeat" — audio replace API

Traditional editors either can't do these tasks or require hand-written scripts (After Effects ExtendScript, DaVinci Lua). OpenCut exposes them over MCP so agents can call them directly. This mirrors projects like ego-lite: don't bolt an agent tool onto the product—design the agent interface into the product architecture.

Rust Core: The Key to Cross-Platform

Video editing is CPU-intensive: real-time preview, multi-track compositing, effects rendering. Traditional cross-platform stacks (Electron, Tauri) bottleneck on video processing. So OpenCut chose a Rust core + native UI per platform:

• Decode, encode, and effects in Rust (high performance, memory safety)
• Native UI layers on desktop, mobile, and browser
• One shared core logic, different UI shells

It resembles Blender's architecture—but Blender is 30 years of accumulated code, while OpenCut starts fresh. Rust's memory safety makes "starting from zero" far cheaper: you don't need 30 years of bug fixes to reach production-grade stability.

Plugin-First: The VS Code Model

The core editor provides minimal functionality; advanced features (effects, transitions, subtitle styles, export formats) all ship as plugins. That's the VS Code playbook, which let it overtake Sublime Text within five years.

Video editing has no true "VS Code" yet: CapCut is closed-source, DaVinci Resolve has a plugin SDK but isn't open, and Kdenlive's plugin ecosystem is weak. Notably, fal.ai—a generative AI platform—has become a sponsor, hinting that AI generation (text-to-video, AI voiceover, AI subtitles) may arrive as plugins.

Headless Mode: The Editor as a Service

Headless mode means the editor can run as a backend service, enabling:

• Automated video pipelines (RSS → video → publish)
• Batch processing (watermarking, transcoding, cutting)
• Video-generation APIs for web apps
• CI/CD integration (commit → auto-generated demo video)

Combined with the MCP server, headless mode makes OpenCut a "programmable video engine"—infrastructure, not just an editor.

The Essential Difference from CapCut

CapCut (ByteDance) is free tool + cloud AI + data collection: your video uploads to their servers for AI processing. OpenCut is MIT-licensed; its architecture keeps video data local—agents operate locally via MCP, and batch processing runs locally via headless mode.

Cloud-first currently dominates because AI features rely on cloud compute. OpenCut bets that local compute (Rust core + GPU) + AI agents (MCP) can beat cloud AI. With increasingly powerful local GPUs (Apple M4, NVIDIA RTX 5090), local 4K processing is already viable, and agents remove the "local means manual" downside.

A Deeper Observation

AI-era tool software isn't "old tools with AI features added"—it's "new tools designed for AI at the architectural level":

• Legacy architecture + AI features: Photoshop generative fill, Premiere auto-edit
• New architecture + AI-native: OpenCut (MCP server + headless + plugin-first)

The difference is like feature phones with browsers versus smartphones. Rewrites are high-risk, but if OpenCut succeeds, it may define not just a better editor but a new architectural paradigm for AI-era tool software.

---

Project: OpenCut-app/OpenCut

Website: opencut.app

Sponsor: fal.ai — generative image, video, and audio model platform