Usage

Skill Library MCP exposes exactly three tools. Two help your assistant find a skill, and one loads it. The whole library is driven through these three:

Tool	Purpose
list_categories	Browse all 13 categories with counts and examples — discover what’s available.
search_skill	Find skills matching a keyword or natural-language query.
load_skill	Load the full content of a specific skill (optionally with its resource files).

Tip

You usually don’t invoke these tools by hand. Describe what you want in plain language — “find me a skill for writing Playwright tests and apply it” — and your assistant decides when to call list_categories, search_skill, and load_skill for you. The examples below show the underlying calls so you understand what’s happening.

The recommended workflow

The three tools are designed to be used in order, narrowing from “what exists” to “the exact skill I want”:

1. list_categories — get the lay of the land. Useful when you don’t yet know what’s available or want to point search at the right area.
2. search_skill — find candidate skills for your task.
3. load_skill — pull in the one you picked.

For a fully worked walkthrough of this loop, see Search and load a skill.

list_categories

Lists every category with its skill count and a few example skills. It takes no parameters. Use it to discover what’s in the library before you search.

list_categories()

The response counts the full library and names each category:

15223 skills in 13 categories:

- **Frontend** (…) — react-patterns, tailwind-layouts, accessibility-audit, +… more
- **Backend** (…) — fastapi-service, graphql-schema, rest-api-design, +… more
- **AI & LLM** (…) — rag-implementation, agent-patterns, prompt-engineering, +… more
…

The 13 categories are: Frontend, Backend, AI & LLM, DevOps & Infra, Data & Databases, Security, Testing, Mobile, Automation, Python, TypeScript & JS, Architecture, and Other (skills that don’t match a keyword category).

search_skill

Searches the library and returns a ranked list of matching skills. Search is IDF-weighted, so it surfaces the most distinctive matches — and it handles natural-language queries, not just bare keywords.

Parameter	Type	Required	Description
query	string	Yes	Keywords or a natural-language description of what you need.

search_skill({ query: "react patterns" })

search_skill({ query: "help me debug a memory leak" })

Each result line shows the skill’s name, its directory name in parentheses, an optional [+resources] marker, and the description:

Found 12 skills matching "react patterns":

- **react-patterns** (react-patterns) [+resources] — Composition, hooks, and render patterns for React apps.
- **react-performance** (react-performance) — Profiling and reducing unnecessary re-renders.
- **state-management** (state-management) — Choosing and structuring client state.
…

Tip

The [+resources] marker means the skill ships extra resources/*.md files beyond its main document. You can pull those in too with include_resources: true when you load it — see below.

If nothing clears the relevance threshold, you get back a plain No skills found matching "<query>". — try broadening the query or running list_categories to find the right area first.

load_skill

Loads the full content of a single skill. This is what actually brings the skill’s instructions into your assistant’s context.

Parameter	Type	Required	Default	Description
name	string	Yes	—	The skill’s frontmatter name or its directory name. Case-insensitive.
include_resources	boolean	No	false	When true, appends the skill’s resources/*.md files to the returned content.

Load just the main skill document:

load_skill({ name: "react-patterns" })

Load it together with its resource files:

load_skill({ name: "react-patterns", include_resources: true })

The name accepts either form shown by search_skill — the human-readable frontmatter name or the directory name — and matching is case-insensitive, so React-Patterns and react-patterns both work.

Caution

include_resources defaults to false. Leave it off for a quick read of the core skill; turn it on only when you want the full deep-dive, since resource files add to what’s loaded into context.

When a name doesn’t match

If load_skill can’t find an exact match, it doesn’t fail silently — it runs a fuzzy search and suggests the closest skills:

load_skill({ name: "react pattern" })
→ Skill "react pattern" not found. Did you mean: react-patterns, react-performance, state-management?

Pick one of the suggestions and call load_skill again with that exact name.

Next steps

• Configuration — install scopes, the plugin-vs-server distinction, and the include_resources toggle.
• Search and load a skill — the full workflow end to end.