home / mcp / claude skills mcp server

Claude Skills MCP Server

MCP server for searching and retrieving Claude Agent Skills using vector search

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "k-dense-ai-claude-skills-mcp": {
      "command": "uvx",
      "args": [
        "claude-skills-mcp"
      ]
    }
  }
}

This MCP server lets any MCP-compatible AI client discover Claude Agent Skills using semantic search and progressive loading. It brings Claude’s Skills framework to multiple AI ecosystems, enabling fast, local discovery of relevant skills from curated repositories and local sources without needing external API keys.

How to use

You access the Claude Skills MCP server from your MCP client and start by loading the claude-skills-mcp backend behind a lightweight frontend. Use the provided MCP entry to connect and begin discovering skills through semantic search. The server offers three core tools you can leverage from your client: find_helpful_skills to locate relevant skills based on a task description, read_skill_document to retrieve specific files from a skill, and list_skills to view the full inventory of loaded skills.

Typical workflow: you describe a task to your MCP client. The client queries the Claude Skills MCP server to find matching skills, loads the most relevant ones, and progressively fetches metadata, full content, then individual files as needed.

How to install

Prerequisites: ensure you can run the MCP frontend and have a compatible runtime environment for the backend. The recommended workflow uses the standalone MCP frontend that automatically downloads the backend.

1) For Cursor users, add the MCP server to your Cursor configuration. You can paste this into your Cursor config file (for example, ~/.cursor/mcp.json):

{
  "mcpServers": {
    "claude-skills": {
      "command": "uvx",
      "args": ["claude-skills-mcp"]
    }
  }
}

The frontend starts instantly and downloads the backend in the background. Expect a short initial wait as the backend loads skill data (roughly 60–120 seconds) before subsequent uses are instant.

2) Run the MCP server in standalone mode with the frontend. Enter this command in your shell to start the server:

uvx claude-skills-mcp

This starts the lightweight frontend which auto-downloads the backend and loads a broad set of Claude skills from official and curated repositories.

3) If you want to customize the configuration, print the default config, modify it, and run with your custom file:

# 1. Print the default configuration
uvx claude-skills-mcp --example-config > config.json

# 2. Edit config.json to your needs

# 3. Run with your custom configuration
uvx claude-skills-mcp --config config.json

Additional configuration and notes

Skills are loaded from multiple sources, including official Claude Skills and curated scientific skills. You can also load from a local directory at ~/.claude/skills if you have relevant content ready for ingestion.

The MCP server supports multiple sources and automatic GitHub caching, with zero need for external API keys. You can connect to a remote backend or run locally as described above.

MCP tools and endpoints

The server exposes three tools for interacting with Claude Agent Skills:

• find_helpful_skills — Semantic search for relevant skills based on a task description.

• read_skill_document — Retrieve specific files (scripts, data, references) from skills.

• list_skills — View the complete inventory of all loaded skills for exploration and debugging.

Notes on architecture

The system uses a two-package architecture: a lightweight frontend and a heavier backend. The frontend starts instantly and downloads the backend in the background, enabling fast interaction and eliminating timeout issues.

The frontend provides a standard MCP server interface for Cursor integration and can connect to a remote backend or run locally with the described commands.

Available tools

find_helpful_skills

Semantic search for relevant Claude Agent Skills based on a task description.

read_skill_document

Retrieve specific files (scripts, data, references) from loaded skills.

list_skills

View the complete inventory of all loaded skills for exploration and debugging.