Claude Code skills vs MCP servers: what's the difference?
Both extend an AI agent, but they solve different problems. A Claude Code skill packages a workflow — the how — as instructions, references, and a test. An MCP server exposes tools and data — the what your agent can call. They're complementary, and Loreto skills can ship with an MCP so you get both.
Claude Code skills: A skill is a folder (SKILL.md + references + test) that teaches your agent a reusable workflow it can run. MCP servers: An MCP server is a running process that exposes tools, resources, and prompts to an agent over the Model Context Protocol.
| Claude Code skills | MCP servers | |
|---|---|---|
| What it provides | A reusable workflow / know-how | Live tools, data, and actions |
| Where it lives | Files in .claude/skills/ | A running server (local or hosted) |
| Best for | Encoding expertise & multi-step processes | Connecting to APIs, databases, systems |
| Runs by | The agent reading & following it | The agent calling its tools |
| Loreto support | Every generated skill is one | Skills can bundle an MCP + CLI |
- You want your agent to repeat a workflow reliably
- The knowledge lives in a video, doc, or your head
- You want something portable you can share or sell
- You need to call a live API, DB, or external system
- You need real-time data or actions, not instructions
The verdict
Use a skill to encode the process; use an MCP to give the agent the tools that process needs. Loreto lets you generate a skill and a matching CLI + MCP from one source.
Related skills
Build the skill for your side of this
Loreto turns any source into a Claude Code skill — package, CLI + MCP, or both.