Skip to content
estudIA

MCP & connectors3 min read

Claude Skills: What They Are and How to Create Your Own

What Claude's Agent Skills are, how they differ from a prompt or an MCP connector, and how to create your own skill for Claude Code or claude.ai.

In short

  1. Pick a repeatable task. Something you do often and always the same way.
  2. Create the folder and SKILL.md. With name and description in the frontmatter and the instructions in the body.
  3. Write a good description. What the skill does and when to use it: it's what Claude reads to decide.
  4. Add resources if needed. Templates, examples or scripts in the same folder.
  5. Install and test it. In .claude/skills/ for Claude Code, or uploaded as a zip in claude.ai settings.

Skills are how you teach Claude to do a task the same way every time without repeating the instructions. A skill is a folder with a SKILL.md file (instructions) and, if needed, templates, examples or scripts. Claude sees a short description of each available skill and only loads the full instructions when the task needs them.

Skill, prompt or connector: what to use

For Lasts
Prompt One specific request One conversation
Project Context and files for one topic The project’s conversations
Skill How to do a type of task, with resources and scripts Always, in any conversation where it fits
MCP connector Access to an app or data While it’s connected

Good examples of skills: “write meeting minutes with our template”, “review a contract against our checklist”, “create slides with the company brand”. Anthropic includes ready-made skills for working with Word, Excel, PowerPoint and PDF documents.

How they work inside

Skills use what Anthropic calls progressive disclosure, an application of context engineering:

  1. Always loaded: only each skill’s name and description (around 100 tokens per skill).
  2. When triggered: the body of SKILL.md with the instructions.
  3. Only if needed: supporting files and scripts. For scripts, only their output enters the context, not their code.

That’s why you can have many skills installed without filling the context window.

Step 1: pick a repeatable task

The best candidate is something you do often, that has a right way of being done and that you’re tired of explaining. If it’s different every time, a prompt is better.

Step 2: create the folder and SKILL.md

A skill for writing minutes, for example, in meeting-minutes/SKILL.md:

---
name: meeting-minutes
description: Writes meeting minutes with the team template from notes or transcripts. Use it when the user asks for minutes, a meeting summary or the tasks agreed in a meeting.
---

# Meeting minutes

## Instructions
1. Identify attendees, date and topic from the material.
2. Summarise decisions in short sentences, one per line.
3. List tasks with owner and date; if one is missing, write "unassigned".
4. Use exactly the structure in template.md.
5. Don't invent anything that isn't in the material.

## Example
See example.md for well-written minutes.

The frontmatter rules, per the official documentation:

  • name: maximum 64 characters, lowercase letters, numbers and hyphens only, and none of the reserved words “anthropic” or “claude”.
  • description: required, maximum 1,024 characters. It must say what it does and when to use it.

Step 3: the description is everything

Claude decides whether to use the skill by matching your request against the description. A vague one (“helps with documents”) won’t trigger, or will trigger at the wrong time. Include the words you’d use when asking (“minutes”, “meeting summary”, “agreed tasks”).

Step 4: add resources

In the same folder you can add:

  • template.md with the exact structure.
  • example.md with a well-solved case.
  • scripts/ with programs for things that must be exact (formatting dates or validating data, for example).

Mention them in SKILL.md so Claude knows when to read them.

Step 5: install it

  • Claude Code: copy the folder to ~/.claude/skills/ (for you, across all your projects) or to .claude/skills/ inside a project (for everyone using that repository). Claude uses it when it fits, and you can also invoke it by typing / followed by its name.
  • claude.ai: zip the folder and upload it under Settings → Features. Available on Pro, Max, Team and Enterprise plans with code execution enabled.
  • API: uploaded through the Skills API and shared across the workspace.

Skills don’t sync between these places: install them in each one.

Safety

A skill can include instructions and code that Claude will run, so Anthropic recommends using only skills from trusted sources (your own or official ones) and thoroughly reviewing any other before installing it, as you would a program. More tips in agent safety.

For inspiration, see Anthropic’s official skills repository. And if what you need is access to an app rather than teaching a task, you want an MCP connector.

Frequently asked questions

What's the difference between a skill and an MCP connector?

A skill teaches Claude how to do something (instructions, templates, scripts). An MCP connector gives it access to what to do it with (an app or data). They complement each other.

Do they sync between Claude Code and claude.ai?

No. Claude Code skills live on your disk, claude.ai skills are uploaded in its settings and API skills are uploaded through the API. You install them in each place.

Glossary terms

Sources

Related articles