Skip to main content
Glama

What are Claude Skills?

Written by on .

mcp
skills
  1. What are Claude Skills?
    1. How Skills Work:
      1. Example Skills include:
      2. Claude Skills vs MCP: Key Differences
        1. When to Use Each
          1. Use Skills when you need:
            1. Use MCP when you need:
            2. The Big Picture
              1. Further Reading

                What are Claude Skills?

                Claude Skills are folders containing instructions, scripts, and resources that Claude can load when needed to perform specific tasks. Introduced by Anthropic in October 2025, they represent a new pattern for extending Claude's capabilities. In December 2025, Anthropic published Agent Skills as an open standard for cross-platform portability.

                How Skills Work:

                • Core format: A Skill is primarily a Markdown file (often named SKILL.md) with YAML frontmatter metadata, optionally accompanied by helper scripts and resources (GitHub repo)

                • On-demand loading: Claude scans available skills at the start of a session, reading only short descriptions from each. The full skill details are only loaded when relevant to the task (Simon Willison)

                • Token efficient: Each skill only takes ~dozens of tokens for its description, with the full content loaded only when needed

                • Code execution required: Skills depend on Claude having access to a coding environment (filesystem access, ability to execute commands)

                Example Skills include:

                • Creating PDFs, Word docs, Excel files, and PowerPoints

                • Making Slack-optimized GIFs

                • Following organization brand guidelines

                • Data processing and visualization tasks

                Skills are available for Claude Pro, Max, Team, and Enterprise plans, as well as Claude Code users and API users with the code execution tool enabled (Claude Help Center).

                Claude Skills vs MCP: Key Differences

                Aspect

                Claude Skills

                MCP (Model Context Protocol)

                What they are

                Markdown files + optional scripts in folders

                A full protocol specification with servers, clients, transports

                Complexity

                Very simple - just text files and scripts

                Complex - covers hosts, clients, servers, resources, prompts, tools, sampling, and multiple transports (stdio, HTTP, SSE)

                Token usage

                Extremely efficient - only loads full details when needed

                Can be token-heavy (GitHub's MCP alone uses tens of thousands of tokens)

                Primary purpose

                Teaching Claude how to perform tasks with procedural knowledge

                Connecting Claude to external data and services

                Requirements

                Requires code execution environment

                Works without code execution; connects to external servers

                Best for

                Workflows, document creation, specialized tasks, internal processes

                Real-time external data access (databases, APIs, third-party services)

                Portability

                Can work with other models (Codex CLI, Gemini CLI)

                Open standard supported by multiple AI platforms

                Ease of creation

                Very easy - write a Markdown file

                More complex - requires building an MCP server

                Sources: MCP In Context, IntuitionLabs, Colin McNamara

                When to Use Each

                Use Skills when you need:

                • To teach Claude a specific workflow or process

                • Document creation (PDFs, spreadsheets, presentations)

                • Tasks that can be accomplished with local code execution

                • Simple, shareable instructions that are easy to iterate on

                • Token-efficient specialized knowledge

                Use MCP when you need:

                • Real-time access to external databases or APIs

                • Connection to third-party services (Slack, GitHub, etc.)

                • Live data that can't be bundled as static files

                • Cross-platform tool integration

                Source: Subramanya.ai

                The Big Picture

                As Simon Willison notes, Skills represent a philosophical shift: "throw in some text and let the model figure it out." They outsource complexity to the LLM and its coding environment rather than requiring formal protocol implementations.

                "MCP is a whole protocol specification, covering hosts, clients, servers, resources, prompts, tools, sampling, roots, elicitation and three different transports. Skills are Markdown with a tiny bit of YAML metadata and some optional scripts... They feel a lot closer to the spirit of LLMs." — Simon Willison

                Some experts suggest Skills and MCP are complementary rather than competitive - you might use Skills for procedural knowledge and MCP for data connectivity (MCP List Blog). In fact, Anthropic has published Skills that can even spin up MCP servers!

                Further Reading

                Written by punkpeye (@punkpeye)