What are Claude AI skills?

Skills are modular capabilities that extend Claude. They consist of instructions, scripts, and resources packaged in a directory starting with a SKILL.md file. Claude discovers and uses them automatically to perform specialized tasks.

What is an MCP server?

MCP (Model Context Protocol) servers are external tools that connect to Claude, giving it access to live data and services like GitHub, Figma, Notion, and databases. They complement skills by providing real-time integrations.

How do I create a Claude skill?

Create a directory with a SKILL.md file that describes the skill's purpose and instructions. Optionally add a scripts folder for automation. Follow Anthropic's skill format and push to GitHub for discovery.

What is the difference between a skill and an MCP server?

Skills provide procedural knowledge and reusable expertise stored locally as SKILL.md files — they require no network connection and persist across conversations. MCP servers connect Claude to external systems and live data via the Model Context Protocol, running continuously to provide real-time access to services like GitHub, Figma, or databases. Skills and MCP servers are complementary: a skill can describe how to analyze data while an MCP server provides the actual data.

How does Claude discover and activate skills?

Claude uses progressive disclosure to manage skills efficiently. It always loads skill metadata (~100 tokens of name and description from YAML frontmatter). When a task matches a skill's description, Claude loads the full SKILL.md instructions. Additional reference files and scripts are only loaded as needed. This means you can bundle comprehensive documentation in a skill without impacting conversation performance.

What should I include in a SKILL.md file?

Every SKILL.md requires a YAML frontmatter block with a name (max 64 characters, lowercase with hyphens) and a description (max 1024 characters). The description is critical — it is the activation trigger that determines when Claude uses the skill. The body should contain procedures, workflows, and context that Claude needs. Keep the file under 500 lines and use separate reference files for advanced content.

Can I use Claude skills without coding?

Yes. Skills are plain Markdown files with a YAML header. No programming knowledge is required to create basic skills. You create a directory, write a SKILL.md file with a description and instructions, and Claude automatically discovers it. Scripts and executables are optional additions for more advanced automation.

SkillHub is a discovery platform for Claude AI skills and MCP servers — a curated, searchable directory similar to how npm is a registry for JavaScript packages. It covers categories including developer tools, testing, DevOps, design, data engineering, and security.

How do I submit a skill to SkillHub?

Visit skillhub.pm/submit to submit your skill. Provide your skill's name, description, GitHub repository URL, and category. Submissions are reviewed by the SkillHub team and, once approved, appear in the public directory. Submission is free.

Are all Claude skills on SkillHub free?

Yes. All skills listed on SkillHub are free to install and use. SkillHub itself is also free for browsing and submitting skills.

Skill Start Guide — Claude AI Skills

Back to Home

Skill Start Guide

Everything you need to know to discover, install, and create skills for Claude AI.

What Are Skills?

Skills are modular, folder-based capabilities that extend what Claude can do. Think of them as expertise packages — collections of instructions, scripts, and resources that Claude automatically discovers and uses when they are relevant to your task.

How Skills Load: Progressive Disclosure

Skills use a three-tier system to stay efficient — only loading what each task requires.

Metadata — Always loaded

~100 tokens. The skill name and description from YAML frontmatter. Claude uses this to decide if a skill is relevant.

Main Instructions — Loaded when triggered

The full SKILL.md body with procedures, workflows, and examples. Only read when the skill becomes relevant.

Additional Resources — Loaded as needed

Reference files, scripts, templates, and data. Zero context cost until accessed.

Key benefit: You can bundle comprehensive documentation without penalty. Claude only loads what each specific task requires — keeping conversations fast and focused.

Skill Structure

Every skill needs at minimum a single SKILL.md file. Additional files are optional and loaded on demand.

File structure

your-skill/
├── SKILL.md          # Required — instructions & metadata
├── reference.md      # Optional — detailed reference docs
├── examples.md       # Optional — usage examples
└── scripts/          # Optional — executable utilities
    ├── analyze.py
    └── validate.py

SKILL.md format

---
name: my-skill-name
description: What this skill does and when to use it
---

# My Skill Name

## Instructions
Your procedures, workflows, and guidance here.

## References
See [reference.md](reference.md) for advanced content.

Name

Max 64 characters. Lowercase letters, numbers, and hyphens only. Use gerund form when possible: processing-pdfs

Description

Max 1024 characters. Must describe both what it does and when to use it. This is the activation mechanism.

Best Practices

Keep it concise

Aim for under 500 lines in SKILL.md. Claude is already smart — only add context it does not have. Use progressive disclosure by referencing separate files for advanced content.

Write effective descriptions

The description determines when Claude activates your skill. Be specific about both purpose and trigger conditions.

Good

"Extract text and tables from PDFs. Use when working with PDF files or document extraction."

Too vague

"Helps with documents"

Match specificity to fragility

Give Claude more or less freedom depending on how sensitive the operation is.

High freedom

Code reviews, analysis, creative tasks — multiple approaches are valid

Medium

Templated tasks with customization — some variation is acceptable

Low freedom

Database migrations, critical sequences — use exact scripts

Test across models

What works for Opus may need more detail for Haiku. Test your skill with all models you plan to use it with: Haiku (needs more guidance), Sonnet (balanced), and Opus (handles less explanation).

Skills Ecosystem

Skills are one part of a broader toolkit. Here is how they compare to related technologies.

Technology	What it does	Persistence	Best for
Skills	Procedural knowledge and reusable expertise	Across conversations	Repeated workflows, domain expertise
MCP Servers	External system access via standardized protocol	Continuous connection	Real-time data, external APIs
Connectors	Pre-built integrations for popular services	Continuous connection	Quick setup for common tools
Prompts	One-time instructions for a single conversation	Single conversation	Ad-hoc tasks, one-off requests
Projects	Background context and reference documents	Project scope	Shared context across team conversations

Skills + MCP are complementary: Skills provide internal procedures and expertise, while MCP connects to external systems. You can combine both for powerful capabilities — for example, a skill that knows how to analyze data paired with an MCP server that provides the data.