The short answer
A Claude Skill is a set of instructions (a SKILL.md file) that teaches Claude a repeatable way to perform a task - your process, format, and standards. MCP is a protocol that connects Claude to external systems - databases, APIs, file systems, and tools - so it can fetch data or take actions.
Side-by-side comparison
| Claude Skills | MCP (Model Context Protocol) | |
|---|---|---|
| What it does | Teaches Claude how to perform a task | Connects Claude to external tools and data |
| Format | A SKILL.md file (markdown + frontmatter) | A server implementing the MCP spec |
| Needs code? | No - plain instructions | Yes - you run/host an MCP server |
| Example | A code-review skill with your standards | A server exposing your database to Claude |
| Who sets it up | Anyone who can write or generate text | A developer who builds or installs a server |
| Best for | Consistent behavior and output | Live data access and actions |
When to use a Claude Skill
- You want Claude to follow a consistent process or output format.
- You're repeating the same instructions in every conversation.
- You don't need live external data - just better, more reliable behavior.
When to use MCP
- Claude needs to read from or write to an external system (database, API, file store).
- You want Claude to take actions in another tool, not just produce text.
- You can run or install a server to expose that capability.
They work together
Skills and MCP aren't competitors - they compose. A skill can describe how to use an MCP-connected tool: the protocol provides access, and the skill provides the playbook. Start with a skill for consistent behavior, and add MCP when Claude needs to reach outside the conversation.
New to skills? Learn how to create one or browse ready-made templates in the Skill Library.