OpenSkills MCP
OpenSkills runs Anthropic's official skills in an isolated container on a Mac and serves them over MCP to any agent: Gemini CLI, Qwen CLI, OpenCode
- 453 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Runs skill code in a container on the local machine and processes user documents and images handed to it by the agent
Install
- Works in
Manual install
chmod +x install.sh && ./install.shNeeds a Mac with Apple Silicon. Starts an MCP server at http://open-skills.local:8222/mcp.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Runs skill code in a container on the local machine and processes user documents and images handed to it by the agent
This is third-party code. Review the repository files before installing.
What it does
OpenSkills installs with the ./install.sh script on a Mac with Apple Silicon and starts an MCP server at http://open-skills.local:8222/mcp. Setup takes about two minutes and runs Apple's native containers, where the skill code actually executes: the user's documents, spreadsheets and images are processed locally, with nothing sent off the machine. The package ships two public skills, image-crop-rotate and pdf-text-replace, each with its own SKILL.md under skills/public/. Any MCP-compatible client connects to the same address: the repository has ready config examples for Claude Desktop, OpenCode, Gemini CLI and Qwen CLI, plus a separate Python client for OpenAI Agents under examples/openai_agents. A user can ask the agent something like "pull every table from these ten PDFs into one Excel file" or "batch process these hundred images: crop to 16:9 and rotate 90 degrees," the agent calls the right skill through MCP, and the code itself runs inside the container on the Mac, not in the cloud. New skills, including Anthropic's own official ones, can be imported and wired into the same server.
Who it is for. For Mac users on Apple Silicon who want to run Anthropic's official skills over sensitive files locally and connect them to more than just Claude, including Gemini CLI, Qwen CLI or another MCP client.
Good fit when
- A financial or personal document needs a skill run on it, but sending it to an external service is off the table: execution happens in a container on the same machine
- You want Anthropic's official skills connected to something other than Claude, such as Gemini CLI or Qwen CLI, through one shared MCP address
- You need bulk batch processing, such as cropping and rotating a hundred images or pulling tables from ten PDFs into one Excel file
- You want to try document and image skills with any MCP client, including OpenCode, without being tied to one specific app
- You want to add your own or a third-party skill to the same local server alongside the bundled image-crop-rotate and pdf-text-replace
Not a fit when
- There is no Mac with Apple Silicon: setup requires macOS and an M1 chip or newer; the README describes no other platform
- You need arbitrary heavy computation inside a skill with no container limits: this is a sandbox for running skills, not a general development environment
- You need a ready catalog of dozens of skills out of the box: the repository ships only two public skills, and the rest need importing yourself
Example request
Crop these hundred images to a 16:9 ratio and rotate them 90 degrees using the image-crop-rotate skill in OpenSkillsMCP
- Transport
- http
- Authentication
- not required
Environment variables
OPENAI_API_KEYsecret- Needed only for the OpenAI Agents example client; other clients work without it.
Limitations
Runs only on a Mac with Apple Silicon and uses Apple's native containers, so install is not documented for Intel Macs, Linux or Windows. Only two skills ship ready to use, image-crop-rotate and pdf-text-replace; everything else needs adding through importing Anthropic's official skills or custom ones. The OpenAI Agents integration needs a separately set OPENAI_API_KEY, so that connection path is no longer fully local.
How to disable. Run cleanup.sh from the repository root; it stops the container and the MCP server. After that, remove the open-skills entry from the MCP client config and delete the repository folder from disk if it is no longer needed.
FAQ
How do I install OpenSkills?
On a Mac with Apple Silicon, clone the repository and run ./install.sh. Setup takes about two minutes and starts an MCP server at http://open-skills.local:8222/mcp.
How do I connect OpenSkills to Gemini CLI or Qwen CLI?
Add http://open-skills.local:8222/mcp to the mcpServers config of those tools; an example for Gemini CLI lives at ~/.gemini/settings.json, with system instructions from examples/gemini_cli/GEMINI.md.
Where do my files go when a skill processes them?
Nowhere off the machine: the skill code runs inside an isolated Apple container on the same Mac where OpenSkills is installed.
What skills are available out of the box?
Two: image-crop-rotate for cropping and rotating images and pdf-text-replace for replacing text in PDFs. Other skills, including Anthropic's official ones, need adding separately.
Can OpenSkills be used without Claude at all?
Yes, the server works with any MCP-compatible client: the repository has examples for OpenCode, Gemini CLI, Qwen CLI and a separate Python client for OpenAI Agents.
Discussion
No questions or reviews yet.
Related
MCP serversSkills Manager keeps skills in one library and mounts them into Claude Code, Cursor, Codex and other agents through symlinks, with no file copies
A library of 229 skills for work and life: sales, marketing, finance, engineering, design and parallel agent runs
A desktop app and skm CLI: one copy of each skill in a central hub, symlinked into every AI tool
A Go CLI that keeps skills, agents and rules in one place and syncs them into every AI agent with one command
Claude Skills (Alireza Rezvani)
Claude Code Skills & Plugins
A large library of skills, agents and commands by domain: engineering, marketing, product, compliance, C-level advisors and research
ai-labs-claude-skills installs 24 Claude skills through one npm package: SEO reports, Docker, CI/CD, CSV analysis, resumes, pitch decks and more
README in short
The README explains the idea of running Claude skills locally on a Mac inside an isolated container and lists the reasons: privacy, full control, compatibility with different MCP clients and the ability to add custom skills. It continues with a quick start, setup for Claude Desktop, OpenCode, Gemini CLI, Qwen CLI and Kiro, a list of example use cases and a short section on adding new skills to the server.