How to Use Higgsfield MCP and CLI to Generate Images and Videos Inside Claude

What Higgsfield CLI actually does

Higgsfield CLI is a command-line layer that connects Higgsfield's image and video generation platform to any AI agent that supports MCP (Model Context Protocol). Once installed, your agent can generate images up to 4K, produce cinematic videos up to 15 seconds, train consistent characters, and pull from your full generation history - all from inside a chat session.

The CLI handles authentication, asset uploads, and async polling automatically. You don't manage API keys or write request code. You describe what you want in plain language, and your agent calls the right Higgsfield model for the job.

It pairs cleanly with Claude Code, Cursor, Codex, OpenClaw, Hermes Agent, and NemoClaw. If your agent supports MCP, Higgsfield connects to it.

Higgsfield MCP or CLI: which one to use

Higgsfield offers two ways in. The Higgsfield MCP is a remote server you connect to chat apps like Claude and ChatGPT, with nothing to install. The Higgsfield CLI is a command-line tool for coding agents like Claude Code, Cursor and Codex. Both use the same account and the same credits.

To use the Higgsfield MCP in Claude, open Claude's settings, go to Connectors, add a custom connector and paste Higgsfield's MCP server URL:

Sign in to your Higgsfield account when Claude asks, then mention @Higgsfield in a chat to generate. If you work in Claude Code, Cursor or Codex instead, follow the CLI install below. Higgsfield's own setup prompt for Claude Code uses the same three commands.

Install in one line

Setup takes three commands. Run them in order and you'll be generating inside your agent within five minutes.

1. 1.Install the CLI globally: npm install -g @higgsfield/cli
2. 2.Authenticate your Higgsfield account: higgsfield auth login — this opens a browser window and completes in about 5 seconds.
3. 3.Add the Higgsfield skills to your agent: npx skills add higgsfield-ai/skills — this syncs the skills registry and makes models like Soul, Seedance, Kling, and Veo available to your agent.

The models you get access to

Once connected, your agent can reach 30+ models across image and video. For images: Soul 2.0 and Nano Banana for photorealistic and stylized stills, Cinema Studio for film-quality renders, Flux 2 and GPT Image 2 for prompt-heavy generation, and Seedream for aesthetic output. For video: Seedance 2.0, Kling 3.0, Veo 3.1, Minimax Hailuo, and WAN 2.6 for motion at up to 15 seconds.

Your agent selects the best model automatically based on your request, or you can name one explicitly. For character-consistent work across multiple images or scenes, Soul Characters handles training and persistence so the same face or product appears correctly across every generation.

What to say to your agent

Once Higgsfield skills are added, you trigger generation with natural language. The prompt structure below works well for most single-asset requests. Adjust model, aspect ratio, and duration based on what you need.

Practical workflows worth knowing

Product photography without a studio: describe your product, the background, and lighting style. Claude calls Nano Banana or Cinema Studio, returns a lifestyle shot ready for listing pages or ads. Iterate by referencing the last output and asking for variations.

Ad creative at scale: describe your campaign angle and target format (UGC, TV spot, unboxing). The marketing skills handle niche research, video generation across formats, and outreach copy. Higgsfield positions this as a replacement for a $5K/mo agency retainer - from one brief.

Content channel automation: feed listings or trending topics into your agent, generate one polished video per item, and distribute via your platform of choice. The full workflow runs from prompt to published asset without leaving the conversation.

Iterative character work: train a Soul Character from reference photos, then generate that character across different scenes, outfits, and locations. Useful for brand ambassadors, influencer content, or consistent visual identities across a campaign.

How credits and pricing work

Higgsfield CLI and the Higgsfield MCP use the same credit balance as the web platform. Each generation costs credits based on the model, length and resolution: a Nano Banana Pro image costs 2 credits, while a 4-second 1080p Veo 3.1 clip costs about 40. There is no separate bill for CLI or MCP usage.

One catch: Higgsfield's Unlimited models and free generations only work on higgsfield.ai, not through the MCP or CLI. Every generation your agent runs spends credits, so pick a plan by credit count. As of October 20, 2026, Basic is $9 a month for 120 credits, Pro is $29 a month ($23 billed annually) for 600 credits and Max is $79 a month ($59 billed annually) for 1,800 credits. Subscription credits do not roll over.

There is no API key to manage. Authentication is tied to your Higgsfield account login, which you complete once via the browser during setup. After that, the CLI handles token refresh in the background.

Before you generate

• Node.js is installed and npm works in your terminal
• Higgsfield CLI installed globally with npm install -g @higgsfield/cli
• Authentication completed with higgsfield auth login
• Skills registry synced with npx skills add higgsfield-ai/skills
• Agent is running (Claude Code, Cursor, Codex, or MCP-compatible client)
• Higgsfield account has sufficient credits for your intended generation

Where this breaks down

Video generation is asynchronous. The agent polls for results, which means there is a wait between request and delivery - anywhere from a few seconds for images to longer for video depending on model and duration. Do not cancel mid-session if the agent appears idle; it is likely waiting on the generation queue.

If your agent does not pick up Higgsfield skills after running the npx command, restart the agent session. Some clients cache the tool registry on startup and will not reflect new additions until relaunched.

Credit errors during generation typically mean your plan does not have enough credits for the requested model or resolution. Check your Higgsfield dashboard and either upgrade or switch to a lighter model.