Higgsfield shipped an official MCP server that exposes 30+ image and video generation models — Soul, Cinema Studio, Flux, Seedream, Kling, Veo and more — through a single hosted endpoint. Once connected, your agent can generate images, create video, train characters and browse history in one session. Here is the setup.
What it can do
- Generate images up to 4K and videos up to ~15 seconds in multiple cinematic styles.
- Train and reuse consistent characters (Soul).
- Pick among 30+ models from one interface.
It works with Claude (web, Cowork, Claude Code), plus other MCP clients.
Setup — official remote server
Higgsfield hosts the server at https://mcp.higgsfield.ai/mcp. Add it as a remote/OAuth connector:
{
"mcpServers": {
"higgsfield": {
"url": "https://mcp.higgsfield.ai/mcp"
}
}
}
Authorise through Higgsfield in the browser on first use. Restart the client; see MCP config file location.
Verify
Ask: "Generate a 4-second cinematic clip of a city at dusk using a Higgsfield video model." A returned video/asset link confirms it works.
Common problems
- Auth/403 — you are not signed in to Higgsfield or your plan lacks generation credits.
- Generations are slow/queued — video models take time; large jobs queue.
- Client lacks remote MCP — use a client that supports HTTP/OAuth servers.
- Credits exhausted — generation consumes plan credits; check your balance.
Cost & safety
Generation uses paid credits, so an over-eager agent can burn through them — keep requests deliberate. Only generate content you have the rights to use. See MCP security best practices.
Going further
Higgsfield pairs with a filesystem server (save outputs) and the Figma or Mobbin servers for a full creative stack. Browse the content category or curated loadouts.