Skillz: MCP server for Claude-style skills
Skillz
An MCP server that opens SKILL.md skills to clients without native support, such as Codex, Copilot and Cursor
- 406 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The server can run helper scripts from skills on your machine
- Skills from third-party sources may contain malicious code, and the author advises running the server in a sandbox or container
Install
- Works in
Manual install
{
"skillz": {
"command": "uvx",
"args": ["skillz@latest"]
}
}An entry for your client's MCP configuration. Skills are read from ~/.skillz.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The server can run helper scripts from skills on your machine
- Skills from third-party sources may contain malicious code, and the author advises running the server in a sandbox or container
This is third-party code. Review the repository files before installing.
What it does
Skillz turns Claude-style skills into MCP tools. You put skills in a folder, the server finds the SKILL.md with a description at the top of each one and exposes the authored instructions to the agent. The rest of a skill's files, meaning scripts, datasets and examples, the agent downloads as resources. The server can also run bundled helper scripts. By default the server looks in the ~/.skillz folder, but you can pass the path to your own directory as the first argument. Skills can sit in a flat layout, in nested subfolders or in archives with the .zip and .skill extensions. It starts over stdio, or over HTTP or SSE with a given host, port and path when needed. The skillz --list-skills command shows which skills the server will see before you connect an agent.
Who it is for. For people who use Codex, Copilot, Cursor or another MCP client and want to plug Anthropic-format skills into it.
Good fit when
- Your agent has no built-in skill support and the skill you need is already written for Claude
- Skills sit in several nested folders or in .zip and .skill archives, and the flat Claude Code layout does not fit
- You want to run the skill server in Docker and mount the skills directory from the host
- One set of skills should serve several MCP clients over HTTP
- You want to check the skill list with --list-skills first and only then connect an agent
Not a fit when
- You work in Claude Code: it reads skills from a folder itself and the bridge is not needed
- Skills come from unverified sources and you are not ready to run the server in a sandbox
- You need to find the right skill by meaning among hundreds: Skillz serves what is in the folder, with no vector search
- You need a stable production tool: the author marks version 0.1 as experimental
Example request
Show which skills are in my ~/.skillz folder and apply the summarize-docs skill to the files in the docs directory.MCP
- Transport
- stdio, http, sse
- Authentication
- not required
Limitations
The author calls the project an experimental proof of concept and warns that it is potentially unsafe: treat skills as untrusted code and run them in a sandbox or container. The PyPI package is marked alpha, with the current version 0.1.14.
Python 3.12 or newer is required. The repository was last updated in January 2026, and the README has no reference for tool names, so check the tool list in the client after connecting.
How to disable. Remove the skillz entry from the MCP client configuration and restart the client. The server does not touch the skills folder, so it stays where it was.
FAQ
How do I connect Skillz to Cursor or Codex?
Add a skillz server to the MCP configuration with the command uvx and the argument skillz@latest. Skills must sit in ~/.skillz. If your folder is different, pass its path as the second argument after skillz@latest.
Can I keep skills in zip archives?
Yes, .zip and .skill archives are supported. Put SKILL.md at the root of the archive or inside its single top-level folder.
Will a Skillz folder work for Claude Code?
Only a flat layout, where each subfolder is one skill. Nested subfolders and archives are read by Skillz alone, and Claude Code skips them.
How do I run Skillz in Docker?
The image is called intellectronica/skillz. Run it with the -i and --rm flags, mount your skills directory into the container as /skillz and pass /skillz as the last argument.
How do I see which skills the server will find?
Run skillz --list-skills, with a path to another directory if needed. The command prints the skills it found and exits, so no agent is needed.
Discussion
No questions or reviews yet.
Related
MCP serversClaude Skills MCP: semantic search for skills
Claude Skills MCP
An MCP server with vector search: the agent describes a task and gets matching skills from the Anthropic and K-Dense collections
A CLI and MCP server for Agent Skills: spec validation, installs from GitHub and on-demand loading of the right skill
A desktop app and CLI to install, sync and organize agent skills across Claude Code, Codex, Cursor and other tools
A universal loader for Anthropic-format skills (SKILL.md) in any coding agent: install, sync and load on demand
The command-line client of the skilld.dev registry that installs agent skills into a project and tracks their changes
A read-only MCP server for the HeyClaude registry: an agent finds skills, MCP servers and hooks and gets ready install commands
README in short
The README gives three ways to run it: through uvx, in Docker and as a Gemini CLI extension. It then shows how the skills directory is laid out, with examples for .zip and .skill, and a table of command line options: transport, host, port, path, skill listing and verbose logging. A link to the Skills Supermarket directory sits at the top.