Configuration

Skill Library MCP is zero-config. There are no environment variables, no API keys, no config file of its own, and no runtime knobs to tune. Run npx -y skill-library-mcp and it indexes its bundled skills on startup and serves them — that’s it.

So this page isn’t a list of settings. It’s an honest rundown of the few decisions you actually make: where you install it, how you install it, and the one toggle that changes behavior at call time.

Tip

If you came here looking for env vars to set, there aren’t any. The server is intentionally self-contained: the skill data ships inside the package and is loaded from disk at launch.

Install scope: user vs. project

When you install through Claude Code, you choose a scope. This decides who sees the server, not how it behaves:

• --scope user — available across all your projects on this machine. This is what the install commands use by default, and it’s the right choice for a general-purpose skill library you’ll want everywhere.
• Project scope — limited to a single project (for example, Cursor’s .cursor/mcp.json or VS Code’s .vscode/mcp.json checked into a repo). Use this when you want the server tied to one codebase, or shared with a team via version control.

Both run the identical server and expose the same three tools — scope only controls where it’s registered.

Plugin vs. MCP server

There are two install shapes for Claude Code, and they’re functionally equivalent:

	Claude Code plugin	MCP server
Install	claude plugin install skill-library	claude mcp add skill-library -- npx -y skill-library-mcp
Server lifecycle	Starts automatically with Claude Code	Registered as a managed MCP server
Tools exposed	search_skill, load_skill, list_categories	Same three
When to pick it	The simple, recommended default	You want to manage the server entry directly

They do the same thing under the hood. Pick one — installing both just gives you two registrations of the same server. See Installation for the exact commands.

The one behavioral toggle: include_resources

The only thing that changes the server’s behavior at runtime is a parameter on load_skill, not a config setting:

• include_resources: false (the default) — load_skill returns just the skill’s main SKILL.md content.
• include_resources: true — it also appends the skill’s resources/*.md files, for the full deep-dive.

This is a per-call choice your assistant makes each time it loads a skill — there’s no global default to flip. See Usage for details.

Where the skill data lives

You don’t manage the skill data — it’s bundled with the package. A skill is simply a directory containing:

• a SKILL.md file with YAML frontmatter (name and description are required), plus the skill’s instructions, and
• an optional resources/ directory of extra .md files that load_skill appends when include_resources: true.

When you run via npx, this data ships inside the published package; nothing is written to your project. If you want to add or modify skills, that’s a manual / local build and a contribution to the source repository, not a configuration step.

Next steps

• Usage — the three tools and the recommended workflow.
• MCP client setup — verify the install and troubleshoot.