Section 1
Why This MCP Matters
The best AI image and video model changes every week. Last month it was Nano Banana. This week it's GPT image 2. Six months ago everyone was rendering with Veo. By the time you read this, that list is already wrong.
Up until now, using each one meant managing a different API, a different account, and a different SDK. Most people just picked one and stuck with it, which means they were almost never using the right tool for the job.
The Higgsfield MCP collapses all of that. One connection inside Claude Code gives you one-shot access to 30+ image and video models, including Soul, Nano Banana, Flux, Seedream, Kling, Minimax Hailuo, Veo, and Higgsfield's own proprietary models.
Once it's wired in, Claude Code stops being a coding assistant and starts being a content production system. You describe what you want, Claude picks the model, generates the asset, brings it back. No tab juggling.
Section 2
Install in Five Minutes
Two install paths. Claude Code (terminal) is the one you want if you plan to automate anything. The claude.ai web connector is fine for quick one-off generation.
Option A: Claude Code (recommended)
Open your terminal in any project. Paste this command:
Install command
$ claude mcp add --transport http -s user higgsfield "https://mcp.higgsfield.ai/mcp" The -s user flag installs Higgsfield globally so it's available in every Claude Code session, not just the project you ran the command in. First time you call a Higgsfield tool, a browser tab opens to log in and authorize. Tell it yes.
Option B: claude.ai web connector
For quick generation in the chat interface (you can see images inline, but you can't automate them):
Open Settings → Connectors
Go to claude.ai, click your profile, then Settings, then Connectors in the left sidebar.
Add Custom Connector
Click Add Custom Connector. Paste the Higgsfield MCP URL: https://mcp.higgsfield.ai/mcp. Hit Add, then Connect.
Log in when prompted
A browser tab opens. Sign in with your Higgsfield account. Done. Higgsfield tools are now available in every claude.ai chat and the desktop app.
You'll need a Higgsfield account. Sign up at higgsfield.ai first. Pricing depends on which models you use. Most image generations cost a few cents per render, video is more.
Section 3
Verify It's Connected
In any Claude Code session, type /mcp. You should see higgsfield listed as connected.
If it's not showing up, fully quit Claude Code and reopen it. New MCP servers only register on a fresh session. If it still doesn't show, run claude mcp list to confirm the install actually landed.
Once connected, ask Claude Code: "Use Higgsfield to generate one test image of a sunrise over mountains using Nano Banana." If a PNG lands in your project folder a minute later, you're live.
One thing about how the MCP works. Higgsfield generations run on their server and don't ping Claude Code when finished. Tell Claude to poll Higgsfield every 60 to 90 seconds and pull the result back. Otherwise it'll wait forever or move on without the image.
Section 4
Eight Use Cases
Each one is a copy-paste prompt for Claude Code. Replace the bracketed placeholders with your own input. Models named in each prompt are the ones I'd reach for first, but you can always tell Claude to swap.
Daily trending carousel
The pattern that's already gone viral once this week. External data source feeds Claude Code, Claude generates the slides through Higgsfield, you post the carousel. Built once, runs every morning.
Pull today's top 5 trending [GitHub repos / X posts / news stories] in [niche]. For each one, write a 30-word body card with a clear hook, the key takeaway, and one specific stat.
Then use the Higgsfield MCP with GPT image 2 (high quality, 2K) to generate:
- 1 cover slide titled "Top 5 [topic] this week" in the same style as ./reference/cover-template.png
- 5 body slides, one per item, in the same style as ./reference/body-template.png
Poll Higgsfield every 60 seconds for completion. Save all 6 PNGs to ./carousel-[YYYY-MM-DD]/. Show me each one as soon as it's ready so I can review before posting. Short-form static ad variants
Generate 6 to 10 image ad variants from one concept and one reference. For Meta, TikTok, or LinkedIn. Same product or hook, different visual angles, so you can A/B test without burning a designer's week.
Use Higgsfield MCP with Nano Banana to generate 8 static ad variants for [PRODUCT/OFFER].
Concept: [ONE-LINE CONCEPT, e.g. "AI assessment that finds 10 hours of busywork in your business"]
Headline overlay: [HEADLINE TEXT]
CTA overlay: [CTA TEXT]
Pull style cues from ./reference/brand-ad-style.png. Format: 1080x1350 (Meta feed) and 1080x1920 (TikTok/Reels). Generate 4 of each.
Variants should differ on: subject angle, color emphasis, text placement, and lifestyle vs product-first framing. Same brand, different hooks.
Poll Higgsfield every 60 seconds. Save to ./ads/[date]/ in two subfolders: feed/ and reel/. Show me thumbnails when ready. Landing page hero images
Stop hunting Unsplash for the hero shot. Generate 4 options that actually fit the page topic, in your brand's color palette, before you decide.
Generate 4 hero image options for a landing page about [PAGE TOPIC, e.g. "an AI assessment that maps your team's busywork in 30 minutes"].
Use Higgsfield MCP with Flux for highest aesthetic quality. 1920x1080.
Style direction: editorial, cinematic, warm color palette (off-white #FAF7F2, warm accent #C4856C, deep ink #1A1A1A). Generous negative space on the left half so headline text can sit there. No human faces. No on-image text.
Vary by: subject (object vs abstract scene vs environment), perspective (wide vs close), and mood (calm vs charged).
Poll every 90 seconds. Save to ./hero-options/ and show me each thumbnail when it lands. Full carousel sets in one batched call
Cover plus 5 to 7 body slides, written and rendered in one shot. The trick is feeding Claude two reference images and locking the style across every output.
Build me a full Instagram carousel on [TOPIC]. Total 7 slides at 1080x1350.
Slide 1 (cover): Big serif title "[CAROUSEL TITLE]" in the same layout as ./reference/cover.png.
Slides 2-7 (body): Each one has a short headline (5 words max) and a 1-2 sentence supporting line, in the same layout as ./reference/body.png.
Outline:
- Slide 2: [POINT 1]
- Slide 3: [POINT 2]
- Slide 4: [POINT 3]
- Slide 5: [POINT 4]
- Slide 6: [POINT 5]
- Slide 7 (CTA): "[CTA TEXT]"
Use Higgsfield MCP with GPT image 2 at high quality. Lock the style by feeding both reference images on every call. Poll every 60 seconds.
Save all 7 slides to ./carousel-[topic-slug]/ numbered 01.png through 07.png. Show me thumbnails as they finish. Hybrid workflow: Higgsfield cover + HTML body
The cost-aware version. Higgsfield only on the slide that needs aesthetics (the cover). Body slides rendered from HTML/CSS using a Playwright headless browser. Same look, fraction of the spend.
Build a 6-slide carousel on [TOPIC] using a hybrid pipeline:
1. Cover slide (slide 1). Use Higgsfield MCP with GPT image 2 at high quality. Match style of ./reference/cover.png. Title: "[COVER TITLE]". 1080x1350.
2. Body slides (slides 2-6). Generate locally using HTML/CSS, no AI image model. Use the template at ./templates/body-slide.html. For each slide, fill in the HEADLINE and BODY slots from the outline below, then render to PNG with Playwright at 1080x1350.
Outline for body slides:
- Slide 2. HEADLINE: "[H2]". BODY: "[B2]".
- Slide 3. HEADLINE: "[H3]". BODY: "[B3]".
- Slide 4. HEADLINE: "[H4]". BODY: "[B4]".
- Slide 5. HEADLINE: "[H5]". BODY: "[B5]".
- Slide 6. HEADLINE: "[CTA HEADLINE]". BODY: "[CTA SUPPORTING LINE]".
Save all 6 to ./carousel-hybrid/ numbered 01.png through 06.png. Tell me the total render cost (Higgsfield only charges for the cover). YouTube and Reel thumbnail variants
Three thumbnail directions per video, before you commit. Different facial framing, different background, different emotional hook, same brand.
Generate 3 thumbnail variants for a [YouTube video / Reel cover] titled "[VIDEO TITLE]".
Use Higgsfield MCP with Nano Banana. Format: [1280x720 for YouTube / 1080x1920 for Reels].
Reference image of me to use for likeness: ./reference/headshot.png. Match facial structure, hair, skin tone exactly.
Variants:
1. Reaction shot: wide-eyed, hand to face, big text overlay reading "[OVERLAY 1]"
2. Confident pose: looking directly at camera, arms folded, text "[OVERLAY 2]" smaller
3. Cinematic: me in profile against a moody backdrop, no overlay text (overlay added later in Figma)
Brand colors: warm off-white background, single warm accent #C4856C, deep ink for text. Poll every 60 seconds. Save to ./thumbnails/ named v1, v2, v3. Studio-style product shots
Feed Higgsfield a real product reference, get back clean studio shots in any setting. Useful for ecom, course covers, software UI marketing, or any branded object you want photographed without booking a photographer.
Generate 4 studio-quality product shots of [PRODUCT NAME] using Higgsfield MCP with Flux.
Reference image of the product: ./reference/product.png. Match the product exactly. Same colors, same logo placement, same proportions.
Settings:
1. Clean studio: white seamless background, soft top-down light, subtle shadow
2. Lifestyle: product on a warm wood desk next to a coffee mug and a notebook, morning light from a window
3. Hero shot: product floating mid-air against a deep neutral background, dramatic side light
4. Editorial: product as one element in a flat-lay with 3-4 complementary objects, top-down view
Format: 1:1 square, 2K. Poll every 60 seconds. Save to ./product-shots/ named studio.png, lifestyle.png, hero.png, editorial.png. B-roll image set for short-form video
Feed Claude your script. It returns a full set of vertical b-roll images, one per beat, all in the same visual style. Drop them into CapCut as 1-2 second cuts. Solves the "I have no b-roll for this video" problem in one prompt.
Read the script at ./scripts/[script-filename].md.
Identify every beat where I would benefit from a b-roll cutaway (visual metaphors, named tools, demonstrations, contrast moments). Aim for 8-12 b-roll moments.
For each beat, write a one-sentence visual description, then use Higgsfield MCP with Nano Banana to generate a 1080x1920 vertical image. All images must share the same visual style: cinematic, slightly desaturated, warm color grade, single subject framing, no on-image text.
Poll every 60 seconds. Save all images to ./broll/[script-name]/ numbered 01.png onward. Output a markdown table mapping each filename to the script line it pairs with so I know where to drop it in CapCut. Section 5
The Daily Automation Pattern
Every use case above is a one-shot generation. The real unlock is when you stop running them manually and turn them into a daily routine. Same shape every time:
External data → Claude analyzes → Builds prompt → Higgsfield renders → Human review
That's the whole framework. Every part is swappable. Pick your data source, pick your output format, run it once a day on a schedule.
Three starter automations to clone. Each one becomes a cron job or a Claude routine that fires every morning before you wake up.
Daily AI news carousel
Source: scrape 3 to 5 X feeds for AI news posts with over 10k engagement in the last 24 hours. Output: 5-slide carousel summarizing the day's top stories. Posts to your Buffer queue at 9am.
Weekly competitor ad rip
Source: pull every static ad your top 3 competitors ran last week from Meta Ad Library. Output: 8 of your own ad variants, same hooks, your branding, ready for upload. Runs every Monday.
Trending sound → matching b-roll generator
Source: top trending audio on TikTok in your niche. Output: 5 to 8 b-roll image sets that match the vibe of each sound, so when you find one you want to use, the visuals are already waiting. Runs every weeknight.
Quick Reference
Everything in One Place
Install (Claude Code)
$ claude mcp add --transport http -s user higgsfield "https://mcp.higgsfield.ai/mcp" Install (claude.ai web)
claude.ai → Settings → Connectors → Add Custom Connector → https://mcp.higgsfield.ai/mcpVerify it's connected
/mcpThe pattern that runs everything
External data → Claude analyzes → Builds prompt → Higgsfield renders → Human review
Models worth knowing
Image: Nano Banana (fast), GPT image 2 (quality), Flux (aesthetic), Soul (proprietary). Video: Kling, Veo, Seedream, Minimax Hailuo. 30+ total. Claude can pick if you don't specify.
Work with Me
Need AI to actually work for your business?
I help businesses cut through the AI hype and build the workflows, automations, and systems that actually move the needle. Direct, hands-on, no fluff.
Work with me