chinavideoai
# China Video AI Prompt MCP
A deterministic, read-only MCP server for designing English or Chinese video prompts, reference-aware shot plans, and focused revisions. It complements [ChinaVideoAI.com](https://chinavideoai.com) and does not generate media, call model providers, access accounts, compare live prices, or spend credits.
## Tools
- `build_video_prompt` turns a rough idea into a structured English or Chinese prompt pack.
- `plan_reference_shots` creates a timed 1–6 shot sequence with reference roles and continuity anchors.
- `diagnose_video_prompt` finds missing motion, camera, lighting, and continuity controls while preserving reference tokens.
- `get_china_video_resources` returns canonical ChinaVideoAI.com guides and workflow pages.
All tools are deterministic and declare read-only MCP annotations. No API key is required.
## Install
Run directly from GitHub:
```json
{
"mcpServers": {
"chinavideoai": {
"command": "npx",
"args": ["-y", "github:gpt-img-2/chinavideoai-prompt-mcp"]
}
}
}
```
Or clone and run locally:
```bash
pnpm install
pnpm build
node dist/index.js
```
Optional environment variable:
- `CHINAVIDEOAI_APP_BASE_URL`: changes the resource-link origin. Defaults to `https://chinavideoai.com`.
## Example inputs
Build a bilingual-aware prompt:
```json
{
"idea": "一只白鹭从清晨薄雾中的湖面起飞",
"workflow": "image-to-video",
"camera": "低机位缓慢跟拍",
"referenceConstraints": "保留 @Image1 中白鹭的羽毛纹理与湖岸构图"
}
```
Plan reference-aware shots:
```json
{
"idea": "A trail shoe crosses wet rock and lands in a shallow stream",
"shotCount": 3,
"totalDurationSeconds": 9,
"referenceRoles": "@Image1 controls product geometry; @Video1 controls motion timing",
"continuityAnchor": "shoe color, laces, runner wardrobe, and travel direction"
}
```
## OpenClaw Skill
The companion Skill is in [`openclaw/china-video-prompt-architect`](openclaw/china-video-prompt-architect/SKILL.md). It works as a text-only workflow without this MCP; connecting the MCP adds deterministic prompt-building and diagnostic tools.
## Development
```bash
pnpm validate
```
This project is an independent prompt-design utility and is not official documentation or an official implementation of Seedance, Kling, Wan, MiniMax, or any other model.
- [中文说明](README.zh-CN.md)
- [China AI video generator](https://chinavideoai.com/china-ai-video-generator)
- [Reference-to-video guide](https://chinavideoai.com/blog/seedance-25-reference-to-video-guide)
## License
MIT
TDQS
Scored across 4 tools
Each tool has a clearly distinct purpose: diagnosing an existing prompt, building a new prompt pack, planning reference shots, and retrieving resources. No two tools could be confused for each other.
All tool names follow the verb_noun pattern with snake_case (diagnose_video_prompt, build_video_prompt, plan_reference_shots, get_china_video_resources). The verbs and nouns are descriptive and consistent.
With 4 tools, the server is well-scoped for its purpose of video prompt creation assistance. Each tool covers a distinct step in the workflow without redundancy.
The toolset covers the full workflow from diagnosis to building to planning, plus resource access. No obvious gaps are apparent for the stated domain of Chinese video prompting.