Atsurae

What 1000ri-jp/atsurae MCP server does

The 1000ri-jp/atsurae MCP server gives an AI agent a structured interface for editing video projects. Its tools cover clip-level timeline work, audio-track management, higher-level timeline operations, visual inspection, quality checks, render control, and rollback. The project describes a five-layer compositing model: background on layer 1, screen captures or slides on layer 2, an avatar on layer 3, effects on layer 4, and text overlays on layer 5.

The server is intended for conversational workflows in which an agent first inspects a project, applies edits, checks the result, and starts a render. The documented output is an MP4 using 1920x1080 resolution, 30fps, H.264 video, and AAC audio.

How it works

The local MCP process acts as the connection between an MCP client and the atsurae.ai service. Tool calls are sent through the MCP protocol to the service's REST API, which performs video-engine operations and FFmpeg rendering. The architecture therefore depends on both the local Python package and access to the configured atsurae.ai API endpoint.

A typical workflow can use atsurae_inspect to view a project's summary, structure, or full timeline, followed by atsurae_edit for clip changes and atsurae_audio for volume, ducking, background music, or narration. atsurae_semantic handles operations such as closing gaps, snapping clips, and reordering them. atsurae_batch can submit up to 20 operations atomically. Preview and analysis tools provide frames, event points, diffs, and checks for gaps, pacing, and composition before rendering.

Setup and configuration

Install the Python package with pip or uv, then run the module locally:

pip install atsurae
python -m atsurae

The 1000ri-jp/atsurae MCP server uses two environment variables in the documented Claude Desktop configuration:

• ATSURAE_API_URL: the atsurae.ai API base URL, shown as https://api.atsurae.ai.
• ATSURAE_API_KEY: the API key used to access that service.

Claude Desktop connects to the local process with the python command and the -m atsurae arguments. After adding the MCP entry to claude_desktop_config.json, restart Claude Desktop before using the tools.

Tools and capabilities

The available tools are:

• atsurae_inspect: inspect project state at summary, structural, or full-timeline detail.
• atsurae_edit: add, move, trim, transform, and delete timeline clips.
• atsurae_audio: manage track volume, ducking, BGM, and narration.
• atsurae_semantic: perform higher-level timeline operations.
• atsurae_batch: execute up to 20 operations as one atomic call.
• atsurae_preview: retrieve preview frames, event points, and before/after differences.
• atsurae_analyze: check for gaps, pacing issues, and composition problems.
• atsurae_render: start, monitor, and download renders.
• atsurae_history: review operations and roll back changes.
• atsurae_pipeline: run the end-to-end AI video creation pipeline.

Limitations and notes

The 1000ri-jp/atsurae MCP server requires an atsurae.ai API URL and API key; the README does not describe an offline-only mode or local rendering configuration. Its documented render format is fixed at 1920x1080, 30fps, H.264 plus AAC in an MP4 container. The public REST API documentation is identified as coming soon, so API-specific behavior beyond the listed MCP tools is not established by the provided material. The repository is MIT licensed.