Higgsfield MCP + Claude — Content Factory Skill Walkthrough (Adil)

Source: raw/Higgsfield_MCP_+_Claude_Just_Changed_Marketing_Forever.md — YouTube, Adil, 2026-05-28.

Walkthrough of a custom Claude Skill (hixfield-content-factory per Adil’s naming, distributed via the video description) that wraps the Higgsfield MCP in a four-stage marketing-content workflow: research → content plan → generate → meta-ads upload. Builds on the existing Higgsfield-MCP tutorials in the topic but adds the load-bearing primitive — a custom Skill that turns Higgsfield from “operator-driven generation” into “agent-driven brand-content factory” runnable from a single Claude chat.

Key Takeaways

• The custom Skill is the load-bearing primitive. Higgsfield MCP alone is generation-on-demand; pairing it with the Content Factory skill turns it into a four-stage pipeline. The skill is what makes the Claude session a marketing agency instead of an image-generation client.
• Four stages, in order:
  1. Research — searches trending content in the configured niche, finds viral references, identifies what’s converting on social media, synthesizes a brief.
  2. Content plan — produces a document with every format, every hook, every platform sequenced by date.
  3. Generation — videos via Higgsfield Marketing Studio, images via GPT Image 2, per-batch permission gates so the operator approves as it goes rather than watching credits disappear post-hoc.
  4. Distribution — connects to Meta Ads and schedules content directly into campaigns. The “Meta Ads upload” closes the loop the prior Higgsfield-MCP tutorials left open.
• MCP install path (~30 seconds). Higgsfield website → MCP and CLI tab → copy connector URL → Claude settings → connectors → “add custom connector” → paste link, name “Higgsfield”, connect to Higgsfield account. Same install path documented in Robo Nuggets’ MCP tutorial.
• Skill install path. Download the skill file from the video description → Claude → customize skills → plus button → create skill → upload skill → drag and drop. Standard custom-skill install.
• Per-batch permission gates as the credit-control discipline. Adil flags this explicitly — the skill is designed so credits get spent in approved batches, not in a runaway generation loop. Pairs with the 50-ad campaign tutorial’s pre-flight balance-check pattern as two ways to keep large generation runs under budget control.
• Demo scope. Adil walks through a 100-video UGC batch run start-to-finish via /pixel content factory invocation. The operator picks the stage to start from (stage 1 in the demo), product URL or image, and the volume target. Claude then handles research → plan → generation → upload across the four stages.
• Where this sits in the topic. Sister to four prior Higgsfield-MCP tutorials in the topic (Mike Futia ad-agency, Robo Nuggets brand-launch, 50-Ad Campaign, Nate Herk creative agency). Adil’s distinctive contribution: the four-stage Skill packaged for one-command operation + meta-ads-upload as the last stage (closing the publish gap none of the prior tutorials addressed end-to-end).
• Skill-as-distribution unit. This article confirms a recurring pattern across recent video-content articles: when a tutorial wraps a multi-tool workflow into a single custom Skill, the Skill itself is the durable artifact, not the prompt sequence. Pairs with Nate Herk’s hypermotion-video skill reverse-engineering pattern.

How it compares to other Higgsfield+Claude tutorials in the topic

Article	Use case	Distinctive contribution
SCALE — DTC campaign	End-to-end ad campaign	Firecrawl brand brief + hero static + UGC clips
Robo Nuggets — brand-launch	Logo to landing page	Side-by-side brand-book (Nano Banana 2 vs GPT Image 2) + 6-panel storyboard + mockup landing
50-Ad Instagram Campaign	Batch ad generation	Playwright Meta Ads Library scraping + 5×5×2 matrix + save-as-skill close
Nate Herk — creative agency	Recurring asset bank	CLI-over-MCP + skill reverse-engineering + Sunday-plan/Monday-generate cron pattern
This article (Adil — Content Factory)	End-to-end brand UGC factory	Four-stage Skill (research → plan → generate → upload) + per-batch permission gates + Meta Ads upload as final stage

The differentiator is the closed-loop publish step. Prior tutorials end at local download or asset bank; Adil’s flow pushes the finished content directly into Meta Ads campaigns.

Related

• AI Video & Content topic
• Higgsfield MCP (connector reference) — the MCP this skill wraps
• skills (Official Bundle) — the vendor-published skill set; this article is a community-authored skill in the same shape
• Higgsfield as a Creative Agency (Nate Herk) — the most architecturally adjacent tutorial (skill-reverse-engineering pattern + recurring-asset framing)
• GPT Image 2 Launch — the image-generation primitive Stage 3 uses
• Meta Ads CLI — adjacent for operators who want CLI-based Meta upload instead of the in-skill connector path

Try It

1. Install Higgsfield MCP via the standard custom-connector path (Higgsfield website → MCP/CLI tab → copy → Claude settings → connectors → add custom).
2. Download the Content Factory skill from Adil’s video description (https://www.youtube.com/watch?v=l7W3QzU8w5s).
3. Install the skill via Claude → customize skills → upload.
4. Run a small dry test first. Invoke /pixel content factory (or whatever the skill registers as) with the smallest video count the skill allows. Walk through the four-stage permission gates, watch where credits get spent, and decide whether to scale.
5. Wire Meta Ads connection before the first large run. The Stage 4 upload step assumes Meta credentials are already connected; verify before kicking off a 100-asset batch.
6. Pair with the 50-Ad Campaign pre-flight balance check — run list_workspaces + balance check before any run >50 assets.

Open Questions

• Skill code visibility. Adil distributes the skill via download link rather than open source repo. Worth fetching and reviewing the SKILL.md before adopting — the prompt patterns inside drive Stage 1-4 routing and quality.
• Creator attribution. Video author is named only as “Adil” without a surfaceable handle in the transcript head. Worth identifying for source-credibility tracking before wider promotion.
• Per-batch permission gates — granularity. “Per batch” isn’t quantified — is one batch one creative concept (5 assets) or one platform (20-50 assets)? The granularity affects how much sleep-time generation is safe.
• Meta Ads upload — actual auth path. “Stage 4 connects to meta ads” is high-level; whether the skill uses Meta Ads CLI underneath, a direct API call, or another MCP path isn’t shown in the head of the transcript. Tail of the transcript would clarify.
• Higgsfield Marketing Studio coverage. This article references Marketing Studio without naming all the modes used. Cross-reference with skills when the modes are enumerated.