Comparisons
MCP and Agent Skills sit next to a lot of similar-sounding ideas. These pages separate them one pair at a time, so you can pick the right layer for the problem in front of you.
The frame behind every comparison
Almost every confusing comparison in this space collapses into one question: are we talking about what an agent can reach, or about what an agent should do? MCP and everything like it answers the first. Skills and everything like them answer the second.
Keep that split in mind and the individual comparisons stop blurring together.
| Layer | Question it answers | Pages in this group |
|---|---|---|
| Access | What can the agent reach? | MCP, MCP server, MCP client, MCP vs API |
| Procedure | What should the agent do? | Skill, SKILL.md, Skill vs prompt |
| Boundary | Which layer owns this concern? | MCP vs Skill, MCP vs function calling |
Start with the comparison you need
The core split. MCP gives access, Skills give procedure, and real workflows use both. Read this one first.
You already have a REST endpoint. Why is that not automatically an MCP server, and when does wrapping it pay off?
Both let a model call tools. One is built into the model's runtime, the other is a protocol across systems.
A prompt is a single instruction. A Skill is a packaged, versioned, activatable procedure. The difference shows up in reliability.
The two halves of one connection. Confusing them causes most beginner implementation mistakes.
Not a comparison but the prerequisite: the definition, the parts, and the request flow.
If you only remember one table
Match your problem to a layer before you pick a format.
Where to go next
If the frame is clear, move from comparison to practice: MCP Quickstart builds one reachable capability, and Skill Quickstart packages one repeatable procedure. To see them used together, read How MCP and Skills work together.