skillz

Skillz is an MCP server that turns Claude-style skills into callable tools for any MCP client. It scans a skills directory (defaulting to ~/.skillz) and exposes each skill along with its described instructions and downloadable resources, allowing agents to discover and invoke skills as tools. You can run Skillz in your environment or via Docker, enabling isolation and easy integration with your MCP clients. To use Skillz from an MCP client, configure the Skillz entry in your MCP configuration so your client can locate and communicate with the Skillz server, and then browse or query the available skills and their resources for use in workflows or agent prompts.

Skillz supports discovering skills packaged as directories or archives (.zip or .skill) containing a SKILL.md front matter and optional resources. Each skill can include scripts, datasets, or examples that agents can download and execute. The server exposes the authored instructions and any resources, and can run bundled helper scripts to assist in executing the skills. You can also run Skillz via Docker for isolation, mounting your skills directory into the container and passing the path to the skills as an argument to the Skillz CLI.

Prerequisites:

• Python 3.8+ (recommended) or a Python-compatible environment
• Access to the PyPI registry to install the Skillz package
• Optional: Docker if you prefer running Skillz in a container

Installation steps:

1. Install Python 3.8+ on your system (from python.org or via your OS package manager).
2. Install Skillz from PyPI:
   • pip install skillz
3. Ensure the UV runner (uvx) is available to launch Skillz (the Quick Start uses uvx to run Skillz). If you don’t already have uvx, install it following its project instructions (commonly via pipx or pip), for example:
   • pipx install uvx or
   • python -m pip install uvx
4. Verify installation by listing available skills (assuming default ~/.skillz directory or your configured skills root) and/or starting Skillz with the recommended command from the README:
   • uvx skillz@latest
5. (Optional) Run Skillz with Docker:
   • docker run -i --rm -v /path/to/skills:/skillz intellectronica/skillz /skillz

If you prefer manual startup without uvx, you can also use equivalent commands provided by your chosen runner after installing Skillz locally.

Tips and common issues:

• Skills are discovered from the root directory you point Skillz at (default: ~/.skillz). Ensure that each skill is either a directory or a packaged archive (.zip or .skill) containing a SKILL.md file with YAML front matter describing the skill, along with any resources.
• When packaging skills, you can nest directories or include archives as shown in the examples; Skillz supports nested layouts and zip/skill packages, unlike some Claude Code setups.
• If you want to verify which skills will be exposed before connecting to an agent, you can run skillz --list-skills (optionally with a custom skills root).
• If you’re using Docker, map your local skills directory into the container (e.g., -v /path/to/skills:/skillz) and pass the mounted path to the Skillz CLI as needed.
• The Skillz server is marked as experimental; run in sandboxed environments if you’re executing untrusted or user-provided skills.