MCP 服务器

Skill MCP Server

A modular plugin system that allows AI agents to dynamically load and use 'skills' from local folders without code changes or restarts. It enables instant capability extension through script execution, hot reloading, and resource bundling.

README

Skill MCP Server 🚀

Turn any AI agent into a specialist — just drop in a skill folder.

<a href="#what-is-skill-mcp-server">📖 What is it?</a> • <a href="#why-choose-skill-mcp-server">🌟 Why Choose It?</a> • <a href="#features">✨ Features</a> • <a href="#quick-start">🚀 Quick Start</a> • <a href="#creating-skills">📝 Creating Skills</a> • <a href="#documentation">📚 Documentation</a>

📖 What is Skill MCP Server?

Skill MCP Server is a standard Model Context Protocol (MCP) server that bridges Claude Skills to any AI agent that supports MCP.

Previously, Claude Skills were mainly used in Anthropic's official tools. If your AI application doesn't support Skills, you'd have to implement your own parsing and execution logic, which is a hassle. With this project, you can simply configure it and let any MCP-compatible Agent use standard Skill files directly.

💡 Core Concepts

🔌 MCP (Model Context Protocol): Think of it as a "USB interface" for AI. As long as your AI assistant supports this interface, it can connect to various tools and services.
📦 Claude Skills: Think of them as "skill packages" for AI. They're not just documentation — they include instructions (SKILL.md), accompanying scripts (Python/JS), and reference materials.

Skill MCP Server is a "converter" that helps various agents use the Skill ecosystem, enabling plug-and-play functionality.

🌟 Why Choose Skill MCP Server?

If your Agent doesn't support Skills yet, this project can help you quickly integrate:

Dimension	Natively Supported Agents (e.g., Claude Code)	Other Agents (with this project)
Access Barrier	Deep integration, usually non-portable	Low barrier, standard MCP protocol
Development Burden	Official implementation complete	Zero code, no need to build Skill parser
Flexibility	Tied to specific clients	Cross-platform, works with any MCP-compatible agent
Feature Parity	Full script, resource & file stream support	Perfect alignment, same dynamic execution & resource access

✨ Features

🛠️ Highly Standardized: Strictly follows MCP protocol
🌍 Universal Compatibility: Not tied to any vendor, works with all MCP-compatible AI clients
⚡ Zero-Code Integration: Helps agents without native Skill support quickly access the Skill ecosystem
📦 Fully Compatible: Supports SKILL.md format and scripts/, references/ resource directories
📂 Workspace Isolation: Supports --workspace parameter to specify where Skill output files are stored
🔄 Hot Reload: Add new skills without restarting the server
🔒 Secure by Design: Path validation, sandboxed file operations

🚀 Quick Start

Recommended: Use uvx to run without manual installation.

📥 Installation

# Using pip
pip install skill-mcp-server

# Using uv (recommended)
uv pip install skill-mcp-server

⚙️ Configure MCP

Add Skill MCP Server to your MCP client configuration. All MCP-compatible clients use the same configuration format:

Using uvx (recommended, no installation needed):

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

Using local installation:

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

Configuration file locations:

Claude Desktop: claude_desktop_config.json (location varies by OS)
Claude Code: ~/.claude.json
Other MCP clients: Refer to your client's documentation

Parameter Explanation:

--skills-dir: Core parameter. Set to the root directory containing all Skill folders you want your agent to use.
--workspace: Important parameter. Specifies where Skill execution output files (code, reports, etc.) are saved.

🛠️ Available Tools (MCP Tools)

Once connected, your AI agent can use the following tools:

🔍 list_skills: List all available skills
📚 skill: Load a specific skill to get detailed instructions from its SKILL.md
📄 skill_resource: Read reference documents or templates from skill packages
▶️ skill_script: Execute scripts bundled with skills in a secure environment
📖 file_read: Read files from the specified workspace
✍️ file_write: Write files to the specified workspace
✏️ file_edit: Edit existing files in the workspace

📝 Creating Skills

A standard Skill structure looks like this:

my-skills/
└── deploy-helper/           # Skill folder
    ├── SKILL.md             # Core documentation (required)
    ├── scripts/             # Executable scripts
    └── references/          # Reference materials

SKILL.md Example:

---
name: deploy-helper
description: Help users deploy applications to production with one click
---

# Deploy Helper Usage Guide

When users request deployment, follow these steps:
1. Use `skill_resource` to read the deployment template.
2. Modify local configuration files.
3. Call `skill_script` to execute the deployment script.

SKILL.md Format

---
name: my-skill
description: Brief description of what this skill does and when to use it
---

# My Skill

## Overview

Explain what this skill enables the AI to do.

## Usage

Step-by-step instructions for the AI agent...

## Available Resources

- `scripts/process_data.py` - Process input data
- `assets/report_template.md` - Output template

💼 Use Cases

📊 Data Analysis: Enable agents to perform data analysis
📝 Document Generation: Enable agents to create professional documents
🔗 API Integration: Enable agents to integrate with specific APIs
🔍 Code Review: Enable agents to follow team standards
🚀 DevOps Tasks: Enable agents to automate deployment workflows

📚 Documentation

🛠️ Development

# Clone the repository
git clone https://github.com/ephemeraldew/skill_mcp.git
cd skill_mcp

# Install development dependencies
uv pip install -e ".[dev]"

# Run tests
pytest

# Run linting
ruff check src/