higgsfield-mcp

这段内容是审核时保存的快照。外部来源才是完整且最新的版本。

---
name: higgsfield-mcp
description: >-
  Creates AI videos and images for TikTok Shop using the Higgsfield MCP connector. Supports Marketing Studio (product URL to video), Soul character training, UGC stills, animated product videos, and batch generation. Use when the user mentions Higgsfield, TikTok Shop videos, UGC creative, AI influencer, Soul training, product video, or marketing video generation.
---

# Higgsfield MCP Skill

## Configuration

- **Auth:** MCP connector — no HTTP calls, no API keys. User must run `/mcp` in Claude Code and select "claude.ai Higgsfield".
- **If tools are unavailable:** Tell the user to run `/mcp` and re-authenticate Higgsfield.
- **Default aspect ratio:** `9:16` (TikTok / vertical)
- **Default mode:** `UGC`
- **Default audio:** `generate_audio: true`

## Available MCP tools

| Tool | Purpose |
|------|---------|
| `generate_video` | Generate a video (Marketing Studio, Seedance 2.0, Kling 3.0) |
| `generate_image` | Generate a still image (Marketing Studio, Soul 2, Nano Banana 2) |
| `job_status` | Poll a generation job until complete |
| `job_display` | Display completed job results in the UI |
| `show_marketing_studio` | Fetch product from URL for Marketing Studio generation |
| `soul_train` | Start training a custom Soul character from reference photos |
| `soul_train_wizard` | Guided Soul training flow |
| `soul_list` | List all trained Souls |
| `soul_status` | Check Soul training progress |
| `media_upload` | Upload reference media (images, videos) |
| `media_confirm` | Confirm uploaded media is ready |
| `show_medias` | Browse media library |
| `models_explore` | Explore available models, aspect ratios, durations |
| `balance` | Check current credit balance |
| `transactions` | View transaction history |
| `show_generations` | Browse past generation history |
| `list_workspaces` | List available workspaces |
| `select_workspace` | Switch active workspace |

## Read order

1. `MASTER_CONTEXT.md` — brand voice, Soul IDs, product defaults, learnings
2. This SKILL.md — tool reference and decision tree
3. `prompting/guide.md` — prompt construction principles
4. Relevant `prompting/prompt-library/*.md` for the chosen workflow

## Decision tree: which flow?

| User goal | Flow | Tool |
|-----------|------|------|
| **Fastest product video** — drop a URL, get a TikTok video | Marketing Studio | `show_marketing_studio` → `generate_video` with `marketing_studio_video` |
| **UGC video with trained Soul** | Soul + Seedance 2.0 | `soul_list` to get Soul ID → `generate_video` with `seedance_2_0` |
| **Train a new AI creator** | Soul training | `media_upload` photos → `soul_train` |
| **Product still / lifestyle image** | Image generation | `generate_image` with `nano_banana_2` or `marketing_studio_image` |
| **UGC still with Soul holding product** | Soul image | `generate_image` with `soul_2` |
| **B-roll / cinematic product motion** | Kling | `generate_video` with `kling3_0` |
| **Batch variations** | Parallel calls | Fire N `generate_video` calls with same params |
| **Check credit balance** | Balance | `balance` |

## Marketing Studio workflow (fastest path)

Use this when the user has a product URL and wants a video fast.

```
1. show_marketing_studio(action="fetch", url=<product_url>, type="webproduct")
   → returns scraping_id and available presets

2. generate_video(
     model="marketing_studio_video",
     url=<product_url>,
     type="webproduct",
     prompt=<TikTok-style script>,
     mode=<preset>,        # UGC | Product Review | Unboxing | Tutorial | Hyper Motion | TV Spot | Wild Card
     aspect_ratio="9:16"
   )

3. job_status(jobId=<id>, sync=true)  ← poll until status="completed"

4. job_display(ids=[<id>])
```

**Preset selection guide:**

| TikTok Shop goal | Recommended preset |
|------------------|--------------------|
| Raw authentic creator review | UGC |
| Feature walkthrough with talking head | Product Review |
| First-open excitement | Unboxing |
| How-to / setup tutorial