Tutorial: Build an AI Marketing Team in Claude Code

Build a Five-Skill AI Marketing Team in Claude Code

By the end of this tutorial, you’ll have a library of five reusable Claude Skills — research, content, creative design, data analysis, and campaign presentation — wired together in Claude Code Desktop and orchestratable through sub-agents or Agent Teams. You’ll build each skill once, then let Claude act as the team lead that deploys them across any campaign task.

1. Create a project folder for your brand with subfolders for /context (brand, product, and audience documents), /data, /examples, and /SOP. Add a CLAUDE.md file at the root — it functions as project-level instructions that every skill in this project inherits automatically, governing how Claude navigates your folder structure and what rules apply across all skills.

1. Install the Skill Creator tool from its official external GitHub repository. The Skill Creator is not bundled with Claude Code — without it, Claude cannot scaffold skill files in the correct structure.

Warning: this step may differ from current official documentation — see the verified version below.

1. Install the Perplexity MCP server. Both the research and social content skills call Perplexity for live web data, so it must be active before you create those skills.

1. In Claude Code Desktop, prompt Claude to create a skill named marketing-research-strategy, referencing your SOP documents as the workflow blueprint and specifying Perplexity MCP as the primary research tool with a fallback. Claude reads the SOP, generates a six-pillar skill file under .claude/skills/marketing-research-strategy/, and makes it invokable via slash command. Calling /marketing-research-strategy analyze [competitor] queues all six research pillars and runs them in parallel — no additional prompting required.

1. Create the social-media-content skill by feeding Claude your best-performing past posts alongside any storytelling frameworks you’ve collected from subject-matter experts. Claude reverse-engineers the patterns and encodes them into the skill, which also calls Perplexity MCP for trending context before drafting copy.

2. Install the Nano Banana MCP server, then prompt Claude to create the social-creative-designer skill. Include your brand’s hex color palette, a default visual style description, and a guard clause that halts execution if Nano Banana MCP is unavailable — this prevents silent failures when image generation is a hard dependency of the skill.

1. Build the data-analysis skill by describing the expected input format and desired dashboard output, then build the campaign-presenter skill by describing how raw campaign materials should transform into a branded presentation or landing page. Both follow the same prompting pattern as the earlier skills.

2. Test multi-skill orchestration with a single natural-language task: generate 10 Instagram posts and matching campaign visuals, then save everything to a new folder. Claude reads brand context, chains the social-media-content and social-creative-designer skills in sequence, and delivers a complete campaign folder — copy, captions, storytelling framework notes, and 20-plus on-brand visuals — without you specifying how to divide the work.

1. For a complex task like a quarterly review, give Claude a single prompt and let it spawn parallel sub-agents — one per workstream. Each sub-agent runs its assigned skill (research, data analysis, or presentation) independently and reports back. Claude synthesizes all outputs into a final deliverable folder without you managing the handoffs.

2. To enable cross-agent review, turn on the Agent Teams feature in global Claude Code settings, define three specialized agents, and assign a task that requires inter-agent feedback before finalization. Token usage is significantly higher than sub-agents, so reserve Agent Teams for workflows where agents reviewing each other’s work materially changes the output quality.

3. Package your skills, agents, and context system into a Claude Plugin with a custom slash command such as /landing-page. This makes the full marketing team portable across projects, clients, and other Claude-compatible environments.

How does this compare to the official docs?

The video moves quickly through installation steps and feature flags that Claude’s official documentation covers in considerably more detail — and in some cases with different defaults than what’s demonstrated here.

Here’s What the Official Docs Show

Act 1 maps a solid end-to-end build you can follow from project setup to campaign delivery. The documentation adds naming precision, environment context, and plan-tier clarity the video moves past quickly — knowing these details upfront prevents the kind of friction that stalls a build mid-step.

1. Create your project folder and context file. The video’s approach here matches the current docs exactly. One naming note worth catching before you start: the official interface surfaces the context file as SKILL.md in the Context panel — the video specifies CLAUDE.md. Both may be accepted as context inputs, but matching the UI convention reduces confusion when cross-referencing community templates or support threads. A clarification that applies to every step that follows: the video refers to the environment as “Claude Code Desktop” throughout; the official product brands the agentic execution mode as Cowork, and Claude Code is described as a terminal, IDE, or browser tool — desktop is one download option, not the exclusive environment.

1. Install the Skill Creator tool from GitHub.

   No official documentation was found for this step — proceed using the video’s approach and verify independently.

2. Install the Perplexity MCP server.

   No official documentation was found for this step — proceed using the video’s approach and verify independently.

3. Create the marketing-research-strategy skill.

   No official documentation was found for this step — proceed using the video’s approach and verify independently.

4. Create the social-media-content skill.

   No official documentation was found for this step — proceed using the video’s approach and verify independently.

5. Install Nano Banana MCP and create the social-creative-designer skill.

   No official documentation was found for this step — proceed using the video’s approach and verify independently.

6. Build the data-analysis and campaign-presenter skills.

   No official documentation was found for this step — proceed using the video’s approach and verify independently.

7. Test multi-skill orchestration with a single prompt. The video’s approach here matches the current docs exactly. Cowork’s numbered Progress panel — visible in the official interface — confirms that Claude natively sequences multi-step tasks, tracks each phase, and allows mid-run cancellation. The Context panel’s native third-party integrations (Notion and Linear are the examples shown) are consistent with the chained-skill pattern the video demonstrates across content and design skills.

1. Spawn parallel sub-agents for a complex task. The video’s approach here matches the current docs exactly — parallel multi-step agentic execution is Cowork’s stated purpose, described officially as letting “Claude power through tasks so you can focus on what matters most.”

1. Enable Agent Teams in global Claude Code settings.

   No official documentation was found for this step — proceed using the video’s approach and verify independently.

2. Package your skills into a Claude Plugin with a custom slash command.

   No official documentation was found for this step — proceed using the video’s approach and verify independently.

One disclosure the video omits entirely: the pricing page confirms a Max plan at $100+/month built for users needing 5–20x more usage than Pro. Running parallel sub-agents, Agent Teams, and multi-skill pipelines at real campaign volume — the workload steps 9–11 describe — is precisely the use case the Max tier is positioned for. Audit your expected usage before committing client workflows to this architecture.

Useful Links

1. Claude — Official Claude Code / Cowork landing page covering the agentic execution environment, Context panel with native third-party integrations, and individual plan pricing including the Max tier for high-volume agentic workloads.