skill-to

skill-to-mcp exposes Claude Skills as MCP resources so you can access them from LLM applications via the Model Context Protocol. The server scans a skills directory recursively for SKILL.md files, parses the YAML frontmatter to extract metadata like skill name and description, and provides three core tools: get_available_skills, get_skill_details, and get_skill_related_file. You can run multiple instances with different skills directories if needed, enabling isolated skill collections for separate projects. To use with an MCP client, configure an mcp.json entry pointing to uvx with the appropriate --skills-dir, and optionally specify UV_PYTHON to pin the Python runtime. Once running, you can query the server for a catalog of skills, fetch detailed skill content, or retrieve specific files within a skill directory as needed.

Prerequisites:

• Python 3.11+ installed on your system
• Access to uvx (via pipx install uvx or via uvx prebuilt)
• A skills directory containing subdirectories with SKILL.md files for each skill

Installation steps:

1. Install Python 3.11+ (https://www.python.org/downloads/)
2. Install uvx (recommended) or ensure it is available in your environment:

# Install uvx via pipx (recommended)
pipx install uvx

3. Prepare your skills directory containing one or more skills, each with a SKILL.md and supporting files
4. Run skill-to-mcp using uvx with the skills directory:

# Example (replace path with your actual skills directory)
UV_PYTHON=3.12 uvx skill_to_mcp --skills-dir /path/to/your/skills

5. Alternatively, embed in an MCP client configuration:

{
  "mcpServers": {
    "skill-to-mcp": {
      "command": "uvx",
      "args": ["skill_to_mcp", "--skills-dir", "/path/to/your/skills"],
      "env": {
        "UV_PYTHON": "3.12"
      }
    }
  }
}

6. For production deployment with environment constraints, set SKILLS_DIR and run:

export SKILLS_DIR=/path/to/your/skills
uvx skill_to_mcp

Tips and common issues: - Ensure SKILLS_DIR (or --skills-dir) points to the directory containing skill subfolders, each with a SKILL.md frontmatter. - If you encounter Python version mismatches, pin UV_PYTHON to the installed runtime version you intend to use. - The three tools (get_available_skills, get_skill_details, get_skill_related_file) are designed to be Security-conscious: the server validates paths to prevent directory traversal outside skill directories. - When integrating with MCP clients, you can run multiple instances with different skills directories to curate separate skill sets per project. - For debugging, enable verbose logs in uvx if available or run with a local skills directory to inspect the SKILL.md content and file lists returned by get_skill_details.