CapcutCli

---

title: capcut-cli type: tool created: 2026-07-16 last_updated: 2026-07-16 related: ["radar/techniques/MCPServerVetting", "radar/techniques/DetectionThenIdentityVideoPipeline", "radar/tools/FireRedOpenStoryline"] sources: ["https://github.com/renezander030/capcut-cli"] radar_quadrant: Tools radar_ring: Assess radar_position: inner

capcut-cli

A command-line tool for editing CapCut and JianYing video projects without the application's interface or any official API. CapCut stores each project as a local JSON draft file; capcut-cli reads and writes that file directly, exposing the timeline to scripts and language-model agents as a deterministic boundary. Published by an independent developer, MIT-licensed, and explicitly not affiliated with ByteDance.

The Problem

CapCut is a graphical editor. Repetitive or bulk edits, such as fixing subtitle timing across dozens of clips, cutting a long recording into shorts, or applying a consistent text style, require manual clicking per project. There is no official automation surface. capcut-cli treats the draft file as the interface, so those operations become one command or one script, and an AI agent can drive them.

Capabilities

Verified against release v0.13.2:

• Timeline repair and mirror-sync for newer CapCut versions where the draft is split across multiple files (sync-timelines, lint --fix)
• Word-level subtitle export to SRT and VTT (export-srt --granularity word)
• Deterministic scene-cut detection (detect-scenes, via ffprobe/ffmpeg)
• Portable text-style presets (make-preset, --preset)
• Keyframe easing native to CapCut (keyframe --easing)
• A queue runner reading JSONL jobs from stdin (capcut serve), for automation platforms such as n8n, Make, and Coze

Distributed both as a global CLI (npm install -g capcut-cli) and as a typed library.

Security and Trust Profile

Assessed at source level on 2026-07-16 against the MCP Server Supply-Chain Vetting gates, which apply equally to any local tool given filesystem access.

• License: MIT, filed correctly. Usage rights are unambiguous.
• Dependencies: none at runtime. The package declares no runtime dependency tree and no install scripts, removing the supply-chain surface that unpinned transitive dependencies create. A global install added a single package.
• Code execution: no dynamic evaluation. External process calls are limited to ffprobe and ffmpeg, built as argument arrays rather than shell strings, and guarded with timeouts.
• Network: no egress during normal edits. Outbound calls exist only in opt-in commands: translation via a user-supplied Anthropic API key, and stock-asset fetches from Wikimedia Commons.
• Safety: every mutating command writes a .bak and a numbered snapshot to a bounded history before the edit; restore --list and restore --step N revert it. Confirmed first-person on 2026-07-16 by adding a text segment and reverting it.

Cross-Platform Note

capcut-cli runs on macOS and Windows. It handles both the CapCut and JianYing draft namespaces and detects newer schema versions rather than assuming a single draft file. Because it operates on the draft JSON directly, correct behaviour on a given machine still depends on that machine's CapCut build using a recognised schema; a read-only inspection (projects, info) confirms compatibility before any write. On 2026-07-16 this was confirmed on macOS: doctor passed and inspection parsed real CapCut 3.3.0 drafts correctly.

Limits

capcut-cli builds and edits the draft; it does not produce the final video. render generates only a low-res ffmpeg proxy preview, and export is an experimental macOS UI-automation path, so the full-quality render stays a manual step in CapCut (command reference). Two headline commands depend on external services the tool does not bundle: translate needs an Anthropic API key and caption needs a Whisper install. JianYing 6.0+ project encryption is unsupported; capcut decrypt only detects it and explains a workaround.

Relationship to Other Blips

capcut-cli is the last stage of a video-editing chain rather than a standalone concern. Detection-then-Identity Video Pipeline (Techniques/Assess/inner) and FireRed-OpenStoryline (Tools/Assess/inner) produce edited footage or storylines; capcut-cli assembles the result into a CapCut project programmatically. Its clean vetting outcome is a concrete instance of MCP Server Supply-Chain Vetting (Techniques/Assess/inner): among four similarly-named CapCut automation projects, star count did not predict trustworthiness, and only source-level review separated the licensed, dependency-free tool from stale forks and networked services.

Radar Assessment

capcut-cli sits in the Assess ring of the Tools quadrant, at inner position. First studied on 2026-07-16 through source-level review, then installed (v0.13.2) and exercised the same day: doctor passed on macOS, read-only inspection parsed the user's real CapCut 3.3.0 drafts, and a mutation-and-undo cycle (add-text then restore) was validated on a throwaway draft. Inner position reflects near-zero adoption cost (single npm install, zero runtime dependencies, MIT licence) and direct applicability to existing video-pipeline work on this radar. It is the strongest of four CapCut automation tools evaluated, the only one that is simultaneously licensed, dependency-free, network-quiet by default, and cross-platform. It remains in Assess rather than Trial because no real project has been edited through to an exported deliverable; the install, schema-recognition, and undo-safety checks that precede that step are now done. The remaining gate for Trial is one of the user's own drafts edited via capcut-cli and rendered to a finished video.

References