AI MediaKit CLI

简体中文

The agent-native command-line toolkit for audio & video. Run Volcengine's cloud AI and local editing through one unified command — built to be driven by AI agents (Claude Code, Trae, Cursor …) or by you in the terminal.

mediakit-cli packs video enhancement, subtitle removal, and a full editing toolbox into a single tool. Heavy AI runs in the cloud; lightweight editing runs locally — switch with one flag, same command surface.

---

✨ What it can do

40+ capabilities across 5 domains — run mediakit-cli --help-full to list them all.

Domain	Capabilities	Runs on	Status
🎬 Editing (17)	trim · concat · watermark · subtitle · speed · volume · filter · flip · fade · mix · mux · extract audio · image-to-video	cloud or local	✅ Available
🎚️ Audio (2)	voice / background separation · audio metadata probe	cloud	✅ Available
🖼️ Image AI (5)	enhance · object / text erase · quality scoring · OCR · background removal	cloud	✅ Available
🎥 Video AI (14)	enhancement (+ generative restore) · subtitle removal · ASR subtitles · OCR · highlight clipping (short-drama / mini-game) · storyline analysis · scene split · portrait & green-screen matting · metadata probe	cloud	✅ Available
🔧 Shared (2)	async task query · remote-file fetch	local / cloud	✅ Available
🚧 Coming	video translation · commentary generation · anime restyling	cloud	Rolling out

AI capabilities run in the cloud (elastic compute, async). Editing runs either in the cloud or locally (sync, zero cost) — pick per command with --cloud / --local.

---

🚀 Quick Start

npm install -g @volcengine/mediakit-cli
npx skills add volcengine/mediakit-cli -g -y   # optional — install agent Skills (Claude Code / Trae / Cursor …)
export MEDIAKIT_API_KEY=<your-api-key>         # from the AI MediaKit console

# Cloud AI (async): enhance to 1080p, then poll for the result
mediakit-cli --cloud video enhance-video --video-url <url> --resolution 1080p
mediakit-cli shared query-task --task-id <task_id> --poll-complete

# Local editing (sync, no key needed): runs on your machine
mediakit-cli --local editing trim-video --video-url ./in.mp4 --start-time 3 --end-time 8

---

📦 Install

# One-click install (CLI + AI agent Skills)
npx @volcengine/mediakit-cli install -y

# npm only (CLI, recommended, cross-platform — pulls the right build for your OS / arch)
npm install -g @volcengine/mediakit-cli

# npx (no install)
npx @volcengine/mediakit-cli version

# curl (macOS / Linux)
curl -fsSL https://raw.githubusercontent.com/volcengine/mediakit-cli/main/scripts/install.sh | bash

Pin a version or path: VERSION=<version> INSTALL_DIR="$HOME/.local/bin" curl -fsSL …/install.sh | bash

Verify: mediakit-cli doctor (checks cloud readiness + local tool deps + install hints).

Update

The CLI checks the npm registry for new releases once a day (TTL 24h). When an update is available you'll see a hint in stderr and a _notice.update field in the stdout JSON.

mediakit-cli version --check       # report current vs latest as JSON
mediakit-cli update --check        # check only, no install
mediakit-cli update                # install the latest via `npm install -g`

Suppress the check with MEDIAKIT_DISABLE_UPDATE_CHECK=1 or in CI (CI env set).

---

🤖 Use with AI Agents

mediakit-cli ships AI agent Skills that teach an agent how to call it — so a user can just say "enhance this video to 1080p and trim the best 5 seconds" and the agent orchestrates the commands.

# One command installs the Skills into every supported agent on your machine
npx skills add volcengine/mediakit-cli -g -y

This auto-detects and installs to 10+ runtimes — Claude Code, Trae (CN & Global), Cursor, Codex, Gemini CLI, GitHub Copilot, OpenCode, OpenClaw, Antigravity, and more.

Every capability is also MCP-compatible — mediakit-cli <domain> <tool> --schema emits a JSON Schema for MCP / Anthropic Tool Use / function-calling, no hand-written adapter needed.

---

🧩 How it works

• Two modes, one command surface. --cloud runs heavy AI in Volcengine's cloud (elastic compute, async task_id); --local runs deterministic editing locally (sync, zero cloud cost). Default mode is cloud-first; per-command flags override it.
• Command structure: mediakit-cli [--cloud|--local] <domain> <tool> [flags] — domains are editing · audio · image · video · shared.
• Outputs: cloud results are returned as URLs; local results write to ~/.mediakit/temp (override with --output-path or MEDIAKIT_OUTPUT_PATH).

---

📖 Documentation

• Volcengine AI MediaKit product docs & pricing: https://www.volcengine.com/docs/6448
• Full command reference & FAQ: see the docs site.
• Error codes & exit code contract — stdout JSON protocol and exit-code rules.

---

🛠 Development

make build          # local build → .mediakit/build/dev/mediakit-cli
make build-all      # all platforms
make snapshot       # snapshot release

Releases are produced via .goreleaser.yml; npm distribution via package.json + scripts/install.js; curl install via scripts/install.sh.

Local Tool Admission (FFmpeg policy)

• ffmpeg / ffprobe: required, 5.1.x, LGPL v2.1 or later, commercial use allowed
• Optional FFmpeg features: openh264, libmp3lame, libass, libfreetype, libfontconfig, libfribidi, libharfbuzz, zlib, libpng, libjpeg-turbo
• Boundary: external process execution only (no static/dynamic linking of local tools into the Go binary); FFmpeg stays in LGPL mode by default; no non-free components; no local intermediate artifacts retained (only final outputs + fetch-file downloads).

---

License

This project is open-sourced under the MIT License.

This software calls MediaKit APIs at runtime. Use of these APIs is subject to the following terms and privacy policies:

• Video Cloud Service Special Terms
• Intelligent Processing Service Billing Rules
• Intelligent Processing Service Level Agreement