skillmake
← marketplace
aitoolsha:26f14bdd52005225manual

higgsfield-mcp

Generate images and videos with Higgsfield's full model roster — Seedance, Kling, Veo, Sora, Soul, Flux, and more — directly from an agent through the official hosted MCP.

Install confidence
curl --create-dirs -fsSL https://skillmake.xyz/i/higgsfield-mcp -o ~/.claude/skills/higgsfield-mcp/SKILL.md
Pinned content
sha:26f14bdd52005225
Generated with
manual
Source
higgsfield.ai

The file served at /api/marketplace/higgsfield-mcp-26f14bdd/raw matches this hash. Inspect before install, then copy the command.

2,767 chars · ~692 tokens
---
name: higgsfield-mcp
description: "Generate images and videos with Higgsfield's full model roster — Seedance, Kling, Veo, Sora, Soul, Flux, and more — directly from an agent through the official hosted MCP."
source: https://higgsfield.ai/mcp
generated: 2026-05-28T19:59:36.155Z
category: tool
audience: ai
---

## When to use

- Generating images via an agent using Higgsfield models (Soul, Nano Banana, Seedream, Flux, GPT Image)
- Generating video from text or image prompts using Seedance, Kling, Veo, Sora, WAN, or Hailuo
- Training and managing character avatars that persist across generations
- Checking credits balance or listing past generations before starting a batch run
- Scoring creative assets for hook strength, retention, and engagement before publishing

## Key concepts

### Hosted OAuth MCP

Higgsfield runs the MCP server at mcp.higgsfield.ai/mcp. No local binary needed — connect once with OAuth and the agent calls tools over HTTP.

### Model routing

The MCP exposes individual tools per model family. Pick the narrowest model that fits the task: image models for stills, video models for motion, character tools for avatar continuity.

### Character training

Upload reference images once to train a character. The trained character ID can then be referenced in subsequent video or image generation calls for consistent identity across outputs.

### Creative analytics

Hook strength, retention score, and engagement scoring tools evaluate finished assets — useful for gating a publish step or comparing variants before human review.

## API reference

```
claude mcp add higgsfield -- npx -y higgsfield-mcp
```

Register Higgsfield as an MCP server using the community npm wrapper (requires HIGGSFIELD_API_KEY in env).

```
claude mcp add higgsfield \
  -e HIGGSFIELD_API_KEY=hf_xxxxxxxxx \
  -- npx -y higgsfield-mcp
```

```
Hosted SSE endpoint: https://mcp.higgsfield.ai/mcp
```

Connect via the official OAuth-authenticated hosted endpoint instead of a local install.

```
Add to mcp config:
{
  "higgsfield": {
    "type": "sse",
    "url": "https://mcp.higgsfield.ai/mcp"
  }
}
```

## Gotchas

- Generated video URLs expire — download assets before referencing them downstream
- Character training is async; poll for completion before using the character ID in a generation call
- Credits are consumed per generation; check balance before running large batches
- Do not expose HIGGSFIELD_API_KEY in logs, screenshots, or generated artifacts
- Workspace context (private vs shared) affects which characters and past generations are visible — set it explicitly when running in a team environment

---
Generated by SkillMake from https://higgsfield.ai/mcp on 2026-05-28T19:59:36.155Z.
Verify against source before relying on details.

File: ~/.claude/skills/higgsfield-mcp/SKILL.md