Skill

Skill MCP Server lets any MCP-compatible AI agent use Claude Skills by acting as a converter layer. It translates standard Skill packages into a format that your agent can execute, enabling plug-and-play access to skills without bespoke parsing or integration work.

How to use

You connect Skill MCP Server to your MCP client so your agent can load and run Skill folders. Once connected, you can quickly access skills, read their documentation, fetch resources, and execute bundled scripts in a sandboxed workspace. This setup supports hot-reloading, so you can add new skills on the fly without restarting the server.

How to install

Prerequisites: ensure you have Python and a runtime for your chosen starting method. You can run the server via uvx for a zero-install experience, or install and run it with Python if you prefer.

# Using uvx (recommended, no local install needed)
uvx
# Then run the Skill MCP Server with your skills directory and workspace via the uvx command shown below

Install the Skill MCP Server package locally if you prefer a Python-based approach.

pip install skill-mcp-server

Additional configuration and usage notes

Configure your MCP client to point to the Skill MCP Server. If you use uvx, provide the server binary and the following arguments to locate your skills and workspace.

{
  "mcpServers": {
    "skill-server": {
      "command": "uvx",
      "args": [
        "skill-mcp-server",
        "--skills-dir", "/path/to/your/skills",
        "--workspace", "/path/to/workspace"
      ]
    }
  }
}

If you install locally with Python, run the server as shown here.

{
  "mcpServers": {
    "skill-server": {
      "command": "python",
      "args": [
        "-m", "skill_mcp_server",
        "--skills-dir", "/path/to/your/skills",
        "--workspace", "/path/to/workspace"
      ]
    }
  }
}

Security and robustness notes

The server validates paths and runs Skill operations in a sandboxed environment. Use the workspace flag to isolate Skill outputs and keep execution results separate from your main project files.

Troubleshooting tips

If Skill loading fails, verify that your skills directory contains valid Skill folders with SKILL.md files and optional scripts and references. Check that the workspace path is writable and accessible by the process running the MCP server.

Notes on usage patterns

Use list_skills to inspect available skills, load a specific skill with skill, read its documentation via skill_resource, and execute associated scripts with skill_script. All file operations respect the designated workspace and are sandboxed for safety.

Available tools

list_skills

List all available skills found in the configured skills directory.

skill

Load a specific skill to retrieve its detailed instructions from the SKILL.md file.

skill_resource

Read reference documents or templates packaged with skills for guidance.

skill_script

Execute bundled scripts from a loaded skill in a controlled environment.

file_read

Read files located in the current workspace to access inputs or outputs.

file_write

Write files to the workspace to persist results and artifacts.

file_edit

Edit existing files within the workspace to modify configurations or outputs.