
Drop-in skill that teaches your agent Higgsfield's motion presets, prompt structure, and render polling — so it stops burning credits on guesswork.
Install command
npm install @scopeful/higgsfield-mcp-betterDownload skill file
higgsfield-mcp.md
5 KB
Fetch via the Scopeful MCP (any client)
Once your agent is connected to the Scopeful MCP, it can load this skill on demand, no install required:
get_skill('higgsfield-mcp-better')Higgsfield ships an MCP server and a CLI. Both work, but agents get the most out of Higgsfield when they understand three things up front: motion presets are the product, prompt structure is opinionated, and renders take longer than image generators. This skill teaches your agent those rules so it stops treating Higgsfield like a generic text-to-video API.
Use Higgsfield when the user wants:
Do not reach for Higgsfield when:
npm install -g @higgsfield/cli
higgsfield auth login
For agent-driven usage, add the Higgsfield MCP server to your agent's MCP config:
{
"mcpServers": {
"higgsfield": {
"command": "npx",
"args": ["-y", "@higgsfield/mcp"]
}
}
}
The MCP server exposes generate_video, list_presets, get_render_status, and download_render.
Higgsfield prompts are not freeform. They have three slots, and skipping any of them produces mediocre output:
Bad: cool video of a car driving fast
Good: red 1970 muscle car parked on wet asphalt, motion: slow push-in, style: rain-soaked neon noir, 35mm anamorphic
When the user gives vague input, ask one clarifying question before generating: which preset? Don't burn a render trying to guess.
Always call list_presets first if you're unsure. Common ones:
push-in / pull-out. distance change, no rotationorbit. rotates around the subjectparallax. sideways camera move with depth separationdolly-zoom. vertigo effect (zoom in, dolly out)whip-pan. fast horizontal sweep, usually for transitionsstatic. no camera motion, use when motion comes from the subjectThe preset is the most important parameter. The same prompt with orbit vs push-in looks like two completely different videos.
Higgsfield renders take 30–120 seconds typically. The MCP generate_video call returns a renderId. Poll get_render_status every 10 seconds, don't poll faster, you'll get rate-limited. Show the user progress, then call download_render once status is complete.
// Typical agent flow
const { renderId } = await mcp.call("higgsfield", "generate_video", {
prompt: "...",
preset: "push-in",
duration: 5,
});
let status = "queued";
while (status !== "complete" && status !== "failed") {
await sleep(10000);
status = (await mcp.call("higgsfield", "get_render_status", { renderId })).status;
}
if (status === "complete") {
const { url } = await mcp.call("higgsfield", "download_render", { renderId });
// Pass `url` to the user.
}
--retry flag does NOT refund the first render, it's a new renderWhen the user asks "how much will this cost?" point them at scopeful.org/tools/higgsfield for the live USD-per-second math.
get_render_status. Higgsfield returns specific codes (NSFW filter, invalid preset, source image too small, etc.)