MCP, MCP 2.0, SKILLS, and Sub-agents
Introduction
If you’ve been learning about AI and Claude, you might have heard terms like MCP, Skills, and Sub-agents. These sound complicated, but they’re really just different ways to give AI assistants more power. Think of them like adding tools to a toolbox — each tool helps the AI do something different.
Let’s break down what each of these is and when you’d use them.
What is MCP (Model Context Protocol)?
MCP stands for Model Context Protocol. It’s a system that lets AI models like Claude connect to tools, databases, and information sources outside of their normal training.
How MCP Works
Imagine Claude is working in a closed room with only what it learned during training. MCP opens a door and lets Claude access:
- Real-time information (current weather, stock prices)
- Company databases (customer information, product details)
- External tools (calculators, code interpreters, email systems)
- Files and documents on your computer
Here’s how it works:
- You set up an MCP server that connects Claude to a tool or data source
- Claude sends a request through MCP asking for information or asking to use a tool
- The MCP server processes the request and sends back the information
- Claude uses that information to answer your question or complete your task
Why MCP is Useful
Without MCP, Claude can only use information from its training data (which has a cutoff date) and what you directly type into the conversation. With MCP, Claude can access live information and perform real actions.
What is MCP 2.0?
MCP 2.0 is the next version of MCP. It’s not completely different — it’s an improvement on the original MCP design.
Key Improvements in MCP 2.0
Stateless Architecture: The big difference in MCP 2.0 is something called “stateless” design. This is a technical term that means each request works independently without needing to remember previous requests.
Think of it like this: Imagine you’re asking a librarian for help. With MCP 1.0, the librarian had to remember everything about you and all your previous requests. With MCP 2.0, the librarian can help you without needing to remember the past — they just look at what you’re asking right now.
This makes MCP 2.0:
- Faster: No need to keep track of history
- More reliable: Each request doesn’t depend on previous ones
- Easier to use: Developers can build MCP servers more simply
When to Use MCP vs MCP 2.0
If you’re building new MCP systems today, you’d use MCP 2.0. If you have old MCP systems that work fine, you don’t need to change them immediately — but MCP 2.0 is the modern standard.
What is a Skill?
A Skill is different from MCP. It’s a custom instruction that tells Claude how to do a specific task or workflow. Think of it like a recipe — it gives step-by-step instructions for accomplishing something.
How Skills Work
A Skill is basically:
- A set of instructions written in a file
- A description of what the Skill does
- Step-by-step directions for Claude to follow
Example: Imagine you have a Skill called “write-tech-news” that does the following:
- Search for tech news from the current week
- Verify stories in multiple sources
- Create a blog post with the stories
- Commit and push the blog post to GitHub
When you ask Claude to use this Skill, it follows all those steps automatically.
Skill vs MCP
Here’s the key difference:
- MCP connects Claude to external tools (databases, APIs, services)
- Skills are instructions that tell Claude how to do something (a workflow or process)
You could think of MCP as giving Claude new powers (like the ability to check a database), and Skills as giving Claude a job description (do this series of steps to accomplish this goal).
What is a Sub-agent?
A Sub-agent is a specialized version of Claude that’s designed to handle one specific type of task. Instead of one general Claude doing everything, you can create specialized Claudes for different jobs.
How Sub-agents Work
When you spawn (create) a Sub-agent, you:
- Tell Claude to start a new instance focused on a specific task
- Give it specialized instructions for that task
- Let it work on the task independently
- It reports back with results
Examples of Sub-agents
Imagine you have these sub-agents:
- Code-Review Agent: Specialized in reviewing code for bugs and improvements
- Research Agent: Specialized in searching the web and finding information
- Deploy Agent: Specialized in deploying applications to servers
- Explore Agent: Specialized in exploring a codebase and finding specific files
Sub-agent vs Skill
The difference is important:
- Skills give Claude instructions for a workflow, but Claude still does the work in the current conversation
- Sub-agents are separate instances of Claude that work independently and report back
Think of it like:
- Skill: “Claude, here are the steps to write a blog post. Please do them.”
- Sub-agent: “I’m creating a specialist Blog-Writer Claude. Please go write that blog post and come back when you’re done.”
Sub-agents are more powerful for complex tasks that require deep focus on one area.
Comparing All Four
| Feature | MCP | MCP 2.0 | Skill | Sub-agent |
|---|---|---|---|---|
| What it is | Protocol for external connections | Better version of MCP | Workflow instructions | Specialized AI instance |
| Purpose | Connect to tools/data | Connect to tools/data | Guide Claude through steps | Handle complex independent tasks |
| Easy to use | Medium | Medium-Easy | Easy | Medium |
| Speed | Depends on tool | Faster than MCP 1.0 | Fast | Slower (separate instance) |
| When to use | Need real-time data | Building modern systems | Simple multi-step tasks | Complex tasks needing focus |
Which Should You Use?
Here’s a quick guide:
Use MCP/MCP 2.0 when:
- You need Claude to access real-time information
- You need Claude to use external tools or services
- You want Claude to work with live databases or files
Use Skills when:
- You have a multi-step process you do repeatedly
- You want to automate a workflow
- You want to keep the process in one conversation
Use Sub-agents when:
- You have a complex task that needs focused work
- You want different specialists for different jobs
- You need to parallelize work (multiple agents working at the same time)
Real-World Example
Let’s say you want to create a weekly blog post about tech news:
- MCP 2.0 — Connects Claude to news APIs to get current articles
- Skill — Gives step-by-step instructions on how to write the blog post
- Sub-agent — You could spawn a specialized “Blog-Writer Agent” to handle the whole process
In reality, you’d probably use all three:
- MCP 2.0 gets the data
- The Skill provides the workflow
- A Sub-agent does the actual work
Final Thoughts
These four systems aren’t really competing — they work together:
- MCP and MCP 2.0 give Claude access to tools and data
- Skills provide structured workflows
- Sub-agents handle complex tasks independently
Understanding what each one does helps you choose the right tool for your problem. As you build more AI projects, you’ll find yourself using all of them in different situations.
The key is to start simple: use what you need, when you need it. You don’t need to use all four at once — but knowing they exist means you have options when your project grows.
Stay curious and keep experimenting!