---
title: "What are Claude skills? SKILL.md playbooks explained"
description: "Claude skills are folders with a SKILL.md file that teach Claude a workflow. How they work, how they differ from MCP servers, and how to install one."
canonical: https://www.leadsets.ai/answers/what-are-claude-skills
updated: 2026-11-09
---

# What are Claude skills?

Claude skills are folders of instructions, and optionally scripts and reference files, that teach Claude how to do a specific task. Each skill has a SKILL.md file with a name and a short description. Claude reads those descriptions and loads the full instructions only when a task calls for it. Skills teach a workflow; tools such as MCP servers give Claude access to data and actions.

|  | Skill | MCP server | Prompt |
| --- | --- | --- | --- |
| What it is | A folder with a `SKILL.md` playbook | A service exposing tools the agent can call | Text you type each time |
| What it gives Claude | Know-how: steps, checks, formats | Access: data, search, actions | One-off instructions |
| Reusable | Yes, loaded when relevant | Yes, always connected | Only if you paste it again |
| Example | Build an outbound prospect list | `https://mcp.leadsets.ai` with 8 tools | "Find me 50 RevOps leads" |

## How Claude skills work

A skill is a small, self-contained package of know-how. At minimum it is a folder with one file, `SKILL.md`, that describes a task and how to do it well. A skill can also include supporting files: templates, reference notes, example outputs or scripts that Claude can run when it has code execution available.

The key idea is that skills load on demand. Claude does not read every skill in full at the start of a conversation. It sees each skill's name and short description, and when your request matches one, it opens the full instructions and any files the skill points to. That keeps the conversation focused: you can have many skills installed without filling Claude's context with instructions it does not need for the current task.

This makes skills a good fit for work you repeat. Instead of re-explaining your process every time ("check the limits first, then search, then verify every email, then split out the uncertain ones"), you write it down once as a skill, and Claude follows it whenever the task comes up.

## What is inside a SKILL.md file

A `SKILL.md` file starts with a short header (YAML frontmatter) that holds the name and description, followed by plain Markdown instructions. The description matters most, because it is what Claude uses to decide when the skill applies.

```
---
name: clean-list-before-sending
description: Verifies every email in a list and splits it into
  send-safe, catch-all, unknown and invalid segments. Use when
  the user asks to clean or check a list before a campaign.
---

# Clean a list before sending

1. Ask for the file and confirm which column holds emails.
2. Verify every address.
3. Return four segments: verified, catch-all, unknown, invalid.
4. Report counts and recommend lower volume for catch-alls.
```

Good skills tend to share a few traits: a specific trigger in the description, numbered steps, explicit checks (what to confirm before moving on), and a defined output format so results are consistent between runs.

## Skills vs MCP servers vs prompts

Skills and MCP servers are often used together, but they solve different problems. An MCP server gives Claude new capabilities: it exposes tools Claude can call, such as searching a database or verifying an email. A skill gives Claude a method: which tools to call, in what order, and how to check the result. If you are new to the protocol side, see [what is an MCP server](https://www.leadsets.ai/answers/what-is-an-mcp-server).

A useful way to think about it: the MCP server is the toolbox, the skill is the playbook, and your prompt is the request. A skill without the right tools can only describe a process. Tools without a skill work, but the agent has to work out the process on its own each time, and results vary more.

Prompts still matter. Skills do not replace telling Claude what you want; they replace the long, repeated instructions about how to do it.

## How to install and use a skill

In the Claude apps, you add a skill by uploading its folder in Claude's settings (the Leadsets library describes this as Claude, Settings, Skills). In Claude Code, skills are folders on disk that Claude picks up from a skills directory. Many skills are shared as Git repositories, so you can clone a whole library at once and keep it up to date.

Once installed, you usually do not call a skill by name. You ask for the outcome in plain language, and Claude chooses the matching skill from its description. If a skill depends on an MCP server, connect that server first; otherwise Claude can read the steps but cannot run them.

Skills are files, so other agents that read the `SKILL.md` format can use them too. The Leadsets [skills library](https://www.leadsets.ai/skills) publishes 50 free, MIT-licensed GTM skills written for Claude, ChatGPT and any agent that reads the SKILL.md standard, available at `github.com/leadsets/leadsets-skills`.

## Example: a prospecting skill in practice

Maya Chen runs revenue operations at Northwind Analytics (northwind.co). She wants Claude to build outbound lists without walking it through the process every week.

She connects the Leadsets MCP server to Claude ([setup steps](https://www.leadsets.ai/docs/mcp)), which gives Claude eight data tools, then installs the [Build an outbound prospect list](https://www.leadsets.ai/skills/build-outbound-prospect-list) skill. She asks: "Build a list of 100 heads of RevOps at US B2B SaaS companies with 50 to 200 people. Put it in a Google Sheet."

The skill tells Claude what to do and in what order: confirm the ICP and ask one question if anything is ambiguous, call `get_usage` to check remaining limits, probe with a one-record `search_companies` call to sanity-check filters, run `search_people` for matching roles, then `enrich_person` and `verify_email` for each person. It returns a table, puts catch-all emails on a separate tab, and reports what it could not confirm.

The skill page shows an illustrative run: 132 matching people, 118 enriched, 96 verified, 14 catch-all and 8 with no email, written to a sheet with three tabs. The value of the skill is not the data itself, which comes from the MCP tools; it is that every run follows the same checks, so Maya gets the same shape of output each time and can see what was uncertain.

## When a skill is worth writing

Write a skill when a task is repeated, has steps that are easy to get wrong, or needs a consistent output. Common examples are report formats, data clean-up routines, code review checklists, brand writing rules and sales workflows such as list building, CRM backfill and pre-send email checks.

A skill is probably not worth it for one-off questions, or for tasks Claude already handles well with a short prompt. Keep skills narrow: one skill per job, with a description that says clearly when to use it. Broad, vague skills are harder for Claude to match to the right request.

Treat skills like code you run. A skill can include scripts, so only install skills from sources you trust, read what they do, and test them on a small input before relying on them.

## Questions and answers

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

An MCP server gives Claude tools it can call, such as searching a database or verifying an email. A skill is a written playbook that tells Claude how to use tools and knowledge to complete a task. They are often used together: the server provides access, the skill provides the method.

### What is a SKILL.md file?

SKILL.md is the main file in a skill folder. It starts with a short header containing the skill's name and description, followed by Markdown instructions. Claude uses the description to decide when the skill applies and reads the instructions when it does.

### Do Claude skills use up context?

Only a little until they are needed. Claude sees each skill's name and description up front and loads the full instructions and files only when a request matches. This lets you install many skills without crowding the conversation.

### How do I install a Claude skill?

In the Claude apps, upload the skill folder in Claude's settings under Skills. In Claude Code, place the skill folder in a skills directory, or clone a skills library from Git. If the skill relies on an MCP server, connect that server as well.

### Do skills work with agents other than Claude?

Skills are plain files, so any agent that reads the SKILL.md format can use them. Leadsets, for example, publishes its 50 GTM skills for Claude, ChatGPT and other agents that follow the SKILL.md standard.

### Are there free Claude skills for sales prospecting?

Yes. Leadsets publishes 50 free, MIT-licensed skills for prospecting, enrichment, email verification, company research, hiring signals, CRM hygiene and outreach handoff. They call Leadsets tools and work on the free key, which includes 1,000 records one time with no card.

---

Leadsets · https://www.leadsets.ai · Site index: https://www.leadsets.ai/llms.txt