Claude Code
for Marketing Teams
AI Agents, Skills & MCPs — Explained Simply
March 12, 2026 • AI Agent World
What We'll Cover
-
Chatbot vs. Assistant vs. AgentUnderstanding the AI landscape
-
MCPs — What Are They?Connecting AI to the real world
-
Skills — Create, Save, Download, UpgradeReusable superpowers for your workflow
-
Agents — Create, Run, CommunicateAutonomous workers at your fingertips
-
Staying Safe — Data, GitHub & Best PracticesProtecting your work and your data
Chatbot vs. Assistant vs. Agent
| Chatbot | Assistant | Agent | |
|---|---|---|---|
| What it does | Answers questions | Answers + follows instructions | Plans, decides, and acts on its own |
| Memory | Remembers within a chat | Remembers your conversation | Remembers + learns your project |
| Can take action? | No — just talks | Limited — with your help | Yes — edits files, runs code, commits |
| Marketing analogy | FAQ page | Smart inbox assistant | Freelancer who does the work |
Claude Code is an agent. It doesn't just answer — it reads your project, makes changes, and verifies the result.
MCPs — What Are They?
MCP = Model Context Protocol. Think of it as a universal adapter that lets Claude plug into external tools.
Without MCPs: Claude can only talk. With MCPs: Claude can check your calendar, draft emails, update your project board — all from one place.
MCP Hosting: Local vs. Cloud
Local MCP Servers
- ▶ Run on your computer
- ▶ Your data stays on your machine
- ▶ You control updates and access
- ▶ Must keep your computer running
Best for: Sensitive data, file access, database work
Cloud / Remote MCP Servers
- ▶ Run on an external server
- ▶ Available 24/7, no setup needed
- ▶ Managed by a third party
- ▶ Data travels over the internet
Best for: Team-shared tools, SaaS integrations, always-on bots
Rule of thumb: If the data is sensitive, host locally. If you need it running all the time or for a team, consider cloud.
Skills — What Are They?
Skills are reusable instructions saved as files. Think of them as recipes that Claude follows every time you need a specific task done.
Documents
"Create a brand brief" — Claude generates a formatted Word doc every time
Reports
"Build a weekly metrics report" — same structure, fresh data
Workflows
"Review a PR and post summary to Slack" — multi-step automation
Skills turn one-time requests into repeatable, shareable, upgradeable capabilities.
Skills: Create, Save, Download, Upgrade
Create a Skill
Ask Claude to do something complex. Then say: "Save this as a skill." Claude writes a SKILL.md file with instructions, inputs, and steps.
Save & Organize
Skills live in your project's .claude/ folder. Name them clearly: "generate-brand-brief", "weekly-report", "social-post-review".
Download & Share
Skills are just text files. Share them with teammates via Git, email, or copy-paste. Anyone with Claude Code can use them.
Upgrade Over Time
Refine your skill: "Add a section for competitor analysis." Claude updates the file. Version control with Git tracks every change.
Skills in Action — Marketing Examples
📝 Campaign Brief Generator
Input: campaign name + goals + audience
Output: formatted brief with timeline, KPIs, messaging framework
📈 Performance Report Builder
Input: raw metrics data
Output: executive summary with charts, insights, and recommendations
💬 Social Post Reviewer
Input: draft social media posts
Output: brand voice check, compliance flags, improvement suggestions
📧 Email Template Factory
Input: campaign type + segment
Output: subject lines, preview text, body copy with A/B variants
Build once, use forever. Each skill gets better as you refine it.
Agents — What Are They?
Agents are autonomous workers that Claude can create to handle complex tasks. They work independently and report back.
"Research competitors"
spawns agents
on sub-tasks
final deliverable
Think of agents like interns: you give them a clear task, they go do the work, and come back with results. Claude manages the team.
Agents: Create, Run & Communicate
Create an Agent
Claude automatically creates agents when a task is complex. You can also ask: "Use an agent to explore our codebase for all API endpoints."
Run Agents
Agents run in the background. Multiple agents can work in parallel — one researches, another drafts, a third reviews. You keep working.
Communicate with Agents
Agents report their findings back to Claude, who combines results and presents them to you. You can ask follow-up questions or redirect.
🔍 Research Agent
"Find all mentions of our brand in the codebase"
🛠 Builder Agent
"Create a landing page based on this wireframe"
How to Not Lose Your Data
Use Git — Always
Git tracks every change. If Claude makes a mistake, you can revert in one command. Think of it as "undo" for your entire project.
Review Before Accepting
Claude shows you every change before applying it. Read the diff. If something looks wrong, say "no" or "undo that."
Use .claudeignore
Create a .claudeignore file to tell Claude to never touch certain files — like credentials, API keys, or sensitive configs.
Permission Modes
"Ask" mode (safest): Claude asks before every action. Start here. Graduate to "auto-accept" once you're comfortable.
Back Up Regularly
Push your Git repository to GitHub. That's your cloud backup. Even if your laptop breaks, your code is safe online.
GitHub in 60 Seconds
GitHub is where your code lives online. It's like Google Drive but for code — with version history built in.
on files locally
of your changes
cloud (backup!)
Creating a Repository
- 1. Go to github.com → "New repository"
- 2. Name it, click Create
- 3. Follow the setup instructions
Uploading Your Code
- 1. Claude can do this for you!
- 2. Ask: "Commit and push to GitHub"
- 3. Or use /commit in Claude Code
Do's and Don'ts
DO
- ✓ Start with small, clear requests
- ✓ Review every change before accepting
- ✓ Use Git to track and revert changes
- ✓ Create a CLAUDE.md for project context
- ✓ Save useful workflows as skills
- ✓ Use /compact during long sessions
- ✓ Push your code to GitHub as backup
DON'T
- ✗ Share your API key with anyone
- ✗ Accept all changes without reading them
- ✗ Paste passwords or secrets into the chat
- ✗ Use "full auto" mode as a beginner
- ✗ Ignore permission prompts
- ✗ Work without Git (no undo = risk!)
- ✗ Skip cost monitoring (/cost)
Your Mission This Week
Complete these four steps before our next session. By the end, you'll have a working agent — and we'll compare everyone's approach to build a better one together.
Open your terminal, run npm install -g @anthropic-ai/claude-code, then run claude and follow the setup prompts.
Navigate to a project folder (or create a new one) and run claude. Have a conversation — ask it to explain the folder, create a file, anything. Get comfortable.
Run claude --remote to get a shareable link. Open it on your phone or tablet — now you can talk to Claude from anywhere, even on the go.
Create an agent that takes a brand name and produces a research dashboard. Design the workflow yourself — what steps, what data, what format? Your thought process is the product.
Everyone's agent will follow a different thought process. We'll compare approaches, find what works best, and combine the best pieces into one ultimate brand research agent — built by the whole team.
Questions?
Let's talk about it.
You'll receive a detailed setup guide
to get Claude Code running on your own device.
docs.anthropic.com • github.com/anthropics/claude-code