atsurae

AIが、あつらえる — AI-crafted video editing

MCP Server for AI-powered video editing. Let Claude, GPT, or any AI agent edit videos through natural language.

Quick Start

hljs language-bash

# Install with pip
pip install atsurae

# Or with uv
uv pip install atsurae

Claude Desktop Configuration

Add to your Claude Desktop config (claude_desktop_config.json):

hljs language-json

{
  "mcpServers": {
    "atsurae": {
      "command": "python",
      "args": ["-m", "atsurae"],
      "env": {
        "ATSURAE_API_URL": "https://api.atsurae.ai",
        "ATSURAE_API_KEY": "your-api-key"
      }
    }
  }
}

Then restart Claude Desktop. You can now edit videos through conversation.

Features — 10 MCP Tools

Tool	Description
`atsurae_inspect`	View project state at 3 detail levels: L1 summary, L2 structure, L3 full timeline
`atsurae_edit`	Add, move, trim, transform, and delete clips on the timeline
`atsurae_audio`	Manage audio tracks — volume, ducking, BGM, narration
`atsurae_semantic`	High-level operations: `close_all_gaps`, `snap_to_previous`, `reorder_clips`
`atsurae_batch`	Execute up to 20 operations atomically in a single call
`atsurae_preview`	Get visual preview frames, event points, and before/after diffs
`atsurae_analyze`	Quality analysis — detect gaps, pacing issues, composition problems
`atsurae_render`	Start, monitor, and download video renders
`atsurae_history`	View operation history and rollback changes
`atsurae_pipeline`	End-to-end AI video creation pipeline

Example

hljs language-vbnet

You: Create a 30-second intro video with the uploaded avatar and background music

Claude: I'll create the intro video. Let me inspect the available assets first...

  [atsurae_inspect] → Found: avatar.mp4, bgm.mp3, logo.png
  [atsurae_edit]    → Placed avatar on Layer 3, logo on Layer 5
  [atsurae_audio]   → Added BGM with -6dB ducking under narration
  [atsurae_semantic] → Closed all gaps, snapped clips
  [atsurae_analyze] → Quality check passed (no gaps, good pacing)
  [atsurae_render]  → Rendering at 1080p...

Claude: Your intro video is ready!
        Duration: 30s | Resolution: 1920x1080 | Size: 12.4 MB
        Download: https://api.atsurae.ai/renders/abc123/output.mp4

Architecture

hljs language-css

                    MCP Protocol                 REST API
[You / AI Agent] ──────────────→ [atsurae MCP] ──────────→ [atsurae.ai API]
                                   Server                        │
                                                                 ▼
                                                          [Video Engine]
                                                                 │
                                                                 ▼
                                                          [FFmpeg Render]
                                                                 │
                                                                 ▼
                                                           [Output MP4]

Layer Compositing Model:

hljs language-vbnet

L5: Telop / Text overlays
L4: Effects (particles, transitions)
L3: Avatar (chroma-keyed)
L2: Screen capture / Slides
L1: Background (3D space, gradients)

Output: 1920x1080, 30fps, H.264 + AAC, MP4

API

atsurae.ai also exposes a REST API that any AI agent can call directly, without MCP.

Documentation: https://docs.atsurae.ai (coming soon)

Development

hljs language-bash

# Clone
git clone https://github.com/1000ri-jp/atsurae.git
cd atsurae

# Install with dev dependencies
uv pip install -e ".[dev]"

# Run the MCP server locally
python -m atsurae

Contributing

Contributions are welcome. Please open an issue first to discuss what you'd like to change.

License

MIT

atsurae is built by 1000ri-jp.

AIが、あつらえる — AI crafts it for you.

atsurae

atsurae

Quick Start

Claude Desktop Configuration

Features — 10 MCP Tools

Example

Architecture

API

Development

Contributing

License

Similar Packages

Tags

Original Author

Publisher