Understanding Claude's Agentic Building Blocks: Skills, Projects, & MCP
🧩 The AI workflow is evolving.
It's no longer just about typing a prompt and getting an answer. Anthropic has introduced a suite of "agentic building blocks" that transform Claude from a chatbot into a powerful workflow engine. But with Skills, Projects, Subagents, and MCP, it can get confusing. Let's break it down.
The 5 Core Building Blocks
Think of building an AI agent like staffing a new department. You need people (Subagents), training manuals (Skills), a filing cabinet (Projects), communication lines (MCP), and daily emails (Prompts).
1. Skills: The "How-To" Manuals
What they are: Reusable sets of instructions that teach Claude how to perform a specific task. They are procedural knowledge.
When to use them: If you find yourself pasting the same "Review this code using these specific security guidelines" prompt over and over, that should be a Skill. Skills load dynamically only when needed, saving your context window.
- Example: A "Brand Voice" skill that teaches Claude your company's specific tone and formatting rules.
2. Projects: The "Knowledge Base"
What they are: Persistent workspaces that hold background knowledge. Everything in a Project is "always on" context.
When to use them: When you have a specific initiative—like "Q4 Marketing Launch"—that requires access to the same set of PDFs, reports, and past chats. It's the "what you need to know" layer.
- Example: A project containing all your competitor analysis reports and product specs.
3. MCP (Model Context Protocol): The "Connectors"
What it is: A universal standard that lets Claude connect to external tools and data sources like Google Drive, Slack, or your local database.
When to use it: When Claude needs to do something outside of the chat window or access live data. If Skills teach Claude how to process data, MCP gives Claude access to the data itself.
- Example: Connecting Claude to your GitHub repo so it can read your actual code files instead of you pasting them in.
4. Subagents: The "Specialists"
What they are: Independent AI assistants with their own specific tools and permissions. They can be delegated tasks by the main agent.
When to use them: For complex workflows where you want separation of concerns. You might have a "Researcher" subagent that can only search the web, and a "Coder" subagent that can write files.
5. Prompts: The "Daily Chat"
What they are: The ephemeral, moment-to-moment instructions you type in the chat box.
When to use them: For one-off requests or to guide the other blocks. "Summarize this" or "Use the Brand Voice skill to rewrite this."
Putting It All Together: The Research Agent
The real magic happens when you combine these blocks. Imagine building a Competitive Analysis Agent:
- 1 Project: You create a "Market Research" project and upload your past year's strategy docs.
- 2 MCP: You connect Google Drive so the agent can pull fresh competitor reports.
- 3 Skills: You add a "SWOT Analysis" skill that teaches Claude exactly how to format the output.
- 4 Prompt: You simply ask: "Analyze our top 3 competitors based on the new reports in Drive."
Claude uses the Project for context, MCP to find the files, the Skill to structure the analysis, and your Prompt to kick it all off.
Ready to start building? Check out our Prompt Vault to start organizing your Skills and Prompts today.
Ready to upgrade your workflow?
Join thousands of power users who trust AI Workspace to organize their prompts and conversations securely.
Install for Free