anime-video-mcp
# anime-video-skill
A Codex Agent Skill and local MCP server for **planning** cited blue-white anime knowledge shorts and an original continuing romance series. The female and male masters supplied with this project are in `references/`.
**Video standard: 1080p, 30fps.** Portrait: 1080×1920; landscape when requested: 1920×1080. H.264/AAC MP4.
**Version 0.3.0:** production planning, local VOICEVOX dialogue (including the repaired Rain v3 male voice), illustrated film/montage scripts, a 30-second HyperFrames director-reference workflow, and a loopback-only ComfyUI video adapter. Both official and minimal MCP servers expose local voice and video tools. Project-wide media placeholder tools remain separate.
The current two-minute film is **《雨》第一章 · 第一幕《初见》**. The new day2 photo scene explicitly uses **720×1280, 30fps, 30 seconds / 900 frames**. Both delivered works are illustrated animatics, without generated continuous body motion or lip synchronization. Generated images, audio, video and weights stay in ignored `outputs/`.
- [Production workflows](docs/production-workflows.md): scripts, dependencies, expected assets, finishing and QA.
- [Local voice](docs/local-voice.md): setup, fixed presets, CLI/MCP and credits.
- [Local video provider](docs/local-video-provider.md): single-shot gate and CPU/MPS diagnostic findings. Wan 5B has not passed; 14B/LTX-2.5 are untested.
- [Six upstream skills](docs/upstream-skills.md) and [pinned paths](config/upstream-skills.lock.json).
- [Day2 script/timeline/Seedance handoff](examples/rain/day2/) and [中文说明](README_zh-CN.md).
Quick check with Python 3.10+:
```bash
python -m pip install -e '.[mcp]'
anime-video episode '为什么夜空是黑的？' --series cosmos --quality draft
anime-video romance-series
anime-video romance-chapter 1
anime-video-mcp
# Optional VOICEVOX runtime required for synthesis (see local voice guide):
anime-video-voice health
anime-video-voice synthesize --speaker male --text 'だから、撮りたくなるんだよ。'
```
The MCP process speaks stdio and does not print a startup banner. The official SDK is preferred; a minimal dependency-free stdio implementation supports offline planning when PyPI is unavailable. The skill is at `skills/one-minute-anime-knowledge/SKILL.md`.
TDQS
Scored across 12 tools
Most tools have clearly distinct names and purposes, but five tools (generate_assets, generate_voice, generate_music, render_episode, qa_episode) share identical descriptions ('Report pending media provider or QA gate'), which could cause misselection if an agent relies solely on descriptions. The distinct names mitigate this, but the overlap in descriptions introduces some ambiguity.
All tool names consistently follow a verb_noun pattern (e.g., create_episode, build_storyboard, generate_voice, qa_episode), with no mixed conventions or style deviations. This makes the naming predictable and easy to navigate.
With 12 tools, the count falls comfortably within the ideal 3-15 range. Each tool represents a distinct step in the anime production pipeline (research, script, storyboard, asset/voice/music generation, render, QA, and batch/series creation), so no tool feels redundant or missing.
The pipeline covers the full lifecycle from research to QA, including script, storyboard, and media generation. Minor gaps exist, such as the absence of explicit update/delete operations or a tool to list existing episodes, but these are not critical for the domain and agents can likely work around them.