MCP Skills Server

Official

by srprasanna · Python

MCP server that loads and serves skills from a mounted directory with hot reload.

This MCP server loads and serves “skills” from a mounted directory and supports hot reload, so changes in the directory can be reflected without restarting the server. It is designed to expose the skills it finds in that location via the Model Context Protocol.

🛠️ Key Features

Loads skills from a mounted directory
Serves skills through an MCP server interface
Hot reload behavior based on directory contents

🚀 Use Cases

Developing or updating a set of MCP skills stored on disk
Running an MCP skills hub backed by a filesystem-mounted directory
Iterating on skills without downtime during reloads

README.md15 KBView on GitHub

MCP Skills Server

A production-ready Model Context Protocol (MCP) server that dynamically loads and exposes skills from a mounted volume with hot-reloading support.

📦 Available on MCP Registry - Install with one command!

Features

Dynamic Skill Loading: Automatically discovers and loads skills from a directory
Hot-Reloading: Detects changes to SKILL.md files and reloads without restart
Folder Structure Validation: Enforces best practices with clear error messages
MCP Protocol Compliant: Full implementation of resources and tools
Production Ready: Comprehensive error handling, logging, and validation
Docker Support: Run in containers with volume mounting
Type Safe: Full type hints using Python 3.13 features
Well Tested: >80% code coverage with comprehensive test suite

Skills Directory Structure
Quick Start
Installation
Usage
Configuration
Skill File Format
MCP Resources and Tools
Development
Troubleshooting
Contributing
License

Skills Directory Structure

CRITICAL REQUIREMENT: Each skill MUST be in its own dedicated folder within the skills directory. The server will ONLY recognize skills that follow this structure.

✅ Valid Structure

code

your-skills-directory/
├── skill-one/
│   └── SKILL.md          ← Required
├── skill-two/
│   ├── SKILL.md          ← Required
│   └── examples/         ← Optional
│       └── example.py
└── skill-three/
    ├── SKILL.md
    ├── examples/
    │   └── demo.py
    └── templates/
        └── template.txt

❌ Invalid Structures (Will Be Ignored)

code

your-skills-directory/
├── SKILL.md                  ❌ Not in a folder - WILL BE SKIPPED
├── random-file.txt           ❌ Not a skill folder
├── .hidden-folder/           ❌ Hidden folder - WILL BE SKIPPED
│   └── SKILL.md
└── __pycache__/              ❌ System folder - WILL BE SKIPPED
    └── SKILL.md

Folder Naming Conventions

Valid folder names:

Lowercase with hyphens: my-skill-name
Lowercase with underscores: excel_advanced
Alphanumeric: skill-name-v2

Invalid (will be skipped):

Hidden folders starting with .
Private folders starting with _
System folders: __pycache__, node_modules, .git, etc.

Quick Start

Using Docker (Recommended)

Create your skills directory:

bash

mkdir -p ~/claude-skills/my-first-skill

Create a skill file:

bash

cat > ~/claude-skills/my-first-skill/SKILL.md << 'EOF'
---
name: "my-first-skill"
description: "My first Claude skill"
---

# My First Skill

This is my first skill for Claude!

## Usage

Simply describe what your skill does here.
EOF

Run the server:

bash

docker run -i --rm \
  -v ~/claude-skills:/skills:ro \
  mcp-skill-hub

Using Poetry (Development)

Clone and install:

bash

git clone https://github.com/srprasanna/mcp-skill-hub.git
cd mcp-skill-hub
poetry install

Create your skills directory:

bash

mkdir -p ~/claude-skills/my-first-skill
# Create SKILL.md as shown above

Run the server:

bash

export MCP_SKILLS_DIR=~/claude-skills
poetry run mcp-skills

Installation

Prerequisites

Python 3.13+ (for development)
Poetry 1.7+ (for dependency management)
Docker (optional, for containerized deployment)

Install with Poetry

bash

# Clone the repository
git clone https://github.com/srprasanna/mcp-skill-hub.git
cd mcp-skill-hub

# Install dependencies
poetry install

# Verify installation
poetry run mcp-skills --help

Build Docker Image

bash

# Build the image
docker build -t mcp-skill-hub .

# Or use docker-compose
docker-compose build

Usage

Running Locally

bash

# Set the skills directory
export MCP_SKILLS_DIR=/path/to/your/skills

# Run the server
poetry run mcp-skills

Running with Docker

bash

docker run -i --rm \
  -v /path/to/your/skills:/skills:ro \
  -e MCP_SKILLS_LOG_LEVEL=INFO \
  mcp-skill-hub

Running with Docker Compose

bash

# Edit docker-compose.yml to set your skills directory path
docker-compose up mcp-skills

Integrating with Claude Desktop

Add to your Claude Desktop configuration (claude_desktop_config.json):

json

{
  "mcpServers": {
    "skills": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-v",
        "${HOME}/claude-skills:/skills:ro",
        "mcp-skill-hub"
      ]
    }
  }
}

Or using Poetry:

json

{
  "mcpServers": {
    "skills": {
      "command": "poetry",
      "args": ["run", "mcp-skills"],
      "cwd": "/path/to/mcp-skill-hub",
      "env": {
        "MCP_SKILLS_DIR": "/path/to/your/skills"
      }
    }
  }
}

Important: Make sure your ${HOME}/claude-skills directory contains skill folders, not loose SKILL.md files!

Configuration

Configuration is done via environment variables with the prefix MCP_SKILLS_:

Variable	Default	Description
`MCP_SKILLS_DIR`	`/skills`	Root directory containing skill folders
`MCP_SKILLS_HOT_RELOAD`	`true`	Enable automatic reloading
`MCP_SKILLS_DEBOUNCE_DELAY`	`0.5`	Delay (seconds) before reload
`MCP_SKILLS_LOG_LEVEL`	`INFO`	Log level (DEBUG, INFO, WARNING, ERROR)
`MCP_SKILLS_SCAN_DEPTH`	`1`	Scan depth (always 1)

Example .env File

bash

MCP_SKILLS_DIR=/path/to/skills
MCP_SKILLS_HOT_RELOAD=true
MCP_SKILLS_DEBOUNCE_DELAY=0.5
MCP_SKILLS_LOG_LEVEL=INFO

Skill File Format

Skills are defined in SKILL.md files with YAML frontmatter:

Minimal Example

markdown

---
name: "my-skill"
description: "Brief description"
---

# My Skill

Your skill content here in Markdown.

Complete Example

markdown

---
# Required fields
name: "excel-advanced"
description: "Advanced Excel automation techniques"

# Version and authorship
version: "1.2.0"
author: "Your Name"
created: "2025-01-15"
updated: "2025-10-23"

# Dependencies
dependencies:
  python: ["openpyxl>=3.0.0", "pandas>=2.0.0"]
  system: ["libreoffice"]

# Categorization
category: "office-automation"
tags: ["excel", "spreadsheet", "automation"]
complexity: "intermediate"  # beginner|intermediate|advanced

# Usage guidance
when_to_use:
  - "Automating Excel report generation"
  - "Processing multiple Excel files"
  - "Creating complex formulas programmatically"

# Relationships
related_skills: ["csv-processing", "data-analysis"]

# Examples
has_examples: true
example_files: ["examples/report_generator.py", "templates/report_template.xlsx"]
---

# Excel Advanced Automation

This skill covers advanced Excel automation techniques...

## Features

- Automated report generation
- Formula creation
- Bulk processing

## Examples

See `examples/report_generator.py` for a working example.

Available Metadata Fields

Required:

name: Unique identifier (kebab-case recommended)
description: Brief description

Optional:

version: Semantic version
author: Creator name
created, updated: ISO dates (YYYY-MM-DD)
dependencies: Python packages or system tools
category: Main category for grouping
tags: Array of tags for search
complexity: beginner, intermediate, or advanced
when_to_use: Array of usage scenarios
related_skills: Names of related skills
has_examples: Boolean flag
example_files: Paths to example files (relative to skill folder)

MCP Resources and Tools

Resources

The server exposes these MCP resources:

skill://catalog - JSON catalog of all skills with metadata
skill://{name} - Individual skill markdown content

Tools

Four tools are available for interacting with skills:

1. `search_skills`

Search for skills by query, category, tag, or complexity.

json

{
  "query": "excel",
  "category": "office-automation",
  "tag": "automation",
  "complexity": "intermediate"
}

2. `reload_skills`

Manually trigger a reload of all skills from the directory.

json

{}

3. `get_skill_info`

Get metadata for a specific skill without loading full content.

json

{
  "name": "excel-advanced"
}

4. `list_skill_folders`

List all valid skill folders found in the skills directory.

json

{}

Development

Setup Development Environment

bash

# Clone repository
git clone https://github.com/srprasanna/mcp-skill-hub.git
cd mcp-skill-hub

# Install dependencies (including dev dependencies)
poetry install

# Activate virtual environment
poetry shell

Running Tests

bash

# Run all tests
poetry run pytest

# Run with coverage
poetry run pytest --cov=mcp_skills --cov-report=html

# Run specific test file
poetry run pytest tests/test_scanner.py

# Run with verbose output
poetry run pytest -v

Code Quality

bash

# Format code
poetry run black .

# Lint code
poetry run ruff check .

# Type checking
poetry run mypy src

# Run all quality checks
poetry run black . && poetry run ruff check . && poetry run mypy src

Development Workflow

Create a branch:
bash
```
git checkout -b feature/my-feature
```
Make changes and test:
bash
```
poetry run pytest
poetry run mypy src
```

Format and lint:

bash

poetry run black .
poetry run ruff check .

Commit and push:

bash

git commit -m "Add feature: description"
git push origin feature/my-feature

Project Structure

code

mcp-skill-hub/
├── src/mcp_skills/          # Source code
│   ├── models/              # Data models
│   ├── parsers/             # Skill parsers
│   ├── storage/             # Repository pattern
│   ├── scanner.py           # Directory scanning
│   ├── watcher.py           # Hot-reload watcher
│   ├── server.py            # MCP server
│   ├── config.py            # Configuration
│   └── __main__.py          # CLI entry point
├── tests/                   # Test suite
├── examples/                # Example skills
├── docs/                    # Documentation
└── pyproject.toml           # Poetry configuration

Troubleshooting

Common Issues

Skills Not Loading

Problem: No skills are loaded when the server starts.

Solution:

Check that your skills are in dedicated folders:

code

/skills/my-skill/SKILL.md  ✓
/skills/SKILL.md           ✗

Verify folder names don't start with . or _
Check logs for detailed error messages

Hot-Reload Not Working

Problem: Changes to SKILL.md files aren't detected.

Solution:

Ensure MCP_SKILLS_HOT_RELOAD=true
Check file is named exactly SKILL.md
Verify file is in a valid skill folder
Look for file watcher errors in logs

Parsing Errors

Problem: SKILL.md files fail to parse.

Solution:

Validate YAML frontmatter syntax
Ensure frontmatter is between --- delimiters
Check required fields (name, description) are present
Use a YAML validator to check syntax

Validation Command

Check your skills directory structure:

bash

poetry run mcp-skills --validate

Expected output:

code

✓ /skills/excel-advanced: Valid skill
✓ /skills/python-automation: Valid skill
✗ /skills/SKILL.md: Error - Skills must be in folders
✗ /skills/.hidden: Skipped - Hidden folder
⚠ /skills/empty-folder: Warning - No SKILL.md found

Summary: 2 valid, 1 error, 1 warning, 1 skipped

Logging

Enable debug logging for detailed information:

bash

export MCP_SKILLS_LOG_LEVEL=DEBUG
poetry run mcp-skills

Logs include:

Folder structure validation messages
Scan progress and results
Parse successes and failures
Hot-reload events
Detailed error context

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

Quick Contributing Guide

Fork the repository
Create a feature branch
Make your changes with tests
Ensure all tests pass and code is formatted
Submit a pull request

Code Standards

Python 3.13+ with type hints
Black for formatting (88 char line length)
Ruff for linting
Mypy for type checking (strict mode)
Pytest for testing (>80% coverage)

Releases

This project uses automated releases via GitHub Actions.

Creating a Release

Go to Actions → Release workflow
Click Run workflow
Choose version bump type:
- patch - Bug fixes (0.1.0 → 0.1.1)
- minor - New features (0.1.0 → 0.2.0)
- major - Breaking changes (0.1.0 → 1.0.0)
- Or specify exact version (e.g., 1.2.3)
Select Docker registry (docker.io or ghcr.io)
Click Run workflow

The workflow will:

✅ Bump version in pyproject.toml
✅ Create Git tag and GitHub release
✅ Build and push Docker image
✅ Run tests to verify release

Docker Images:

Docker Hub: {username}/mcp-skill-hub:{version}
GitHub: ghcr.io/{owner}/mcp-skill-hub:{version}

See RELEASING.md for detailed release documentation.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

Built with MCP Python SDK
Inspired by the need for dynamic skill management in Claude
Thanks to all contributors!

Note: This server makes it impossible to misunderstand the folder structure requirement through:

Clear error messages with folder context
Comprehensive logging
Validation at multiple levels
Detailed documentation
Working examples

Each skill MUST be in its own folder. This design decision ensures clean organization, easy management, and unambiguous structure. 🎯

Install

Configuration

Environment variables

MCP_SKILLS_DIRdefault /skills

Root directory containing skill folders

MCP_SKILLS_HOT_RELOADdefault true

Enable automatic reloading when SKILL.md files change

MCP_SKILLS_LOG_LEVELdefault INFO

Logging verbosity: DEBUG, INFO, WARNING, ERROR, CRITICAL

Add to VS Code Add to Cursor

claude_desktop_config.json

json

{
  "mcpServers": {
    "mcp-skill-hub": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "docker.io/srprasanna/mcp-skill-hub:1.0.0"
      ],
      "env": {
        "MCP_SKILLS_DIR": "/skills",
        "MCP_SKILLS_HOT_RELOAD": "true",
        "MCP_SKILLS_LOG_LEVEL": "INFO"
      }
    }
  }
}

🛠️ Key Features

🚀 Use Cases

MCP Skills Server

Features

Table of Contents

Skills Directory Structure

✅ Valid Structure

❌ Invalid Structures (Will Be Ignored)

Folder Naming Conventions

Quick Start

Using Docker (Recommended)

Using Poetry (Development)

Installation

Prerequisites

Install with Poetry

Build Docker Image

Usage

Running Locally

Running with Docker

Running with Docker Compose

Integrating with Claude Desktop

Configuration

Example .env File

Skill File Format

Minimal Example

Complete Example

Available Metadata Fields

MCP Resources and Tools

Resources

Tools

1. search_skills

2. reload_skills

3. get_skill_info

4. list_skill_folders

Development

Setup Development Environment

Running Tests

Code Quality

Development Workflow

Project Structure

Troubleshooting

Common Issues

Skills Not Loading

Hot-Reload Not Working

Parsing Errors

Validation Command

Logging

Contributing

Quick Contributing Guide

Code Standards

Releases

Creating a Release

License

Acknowledgments

Install

Configuration

Related servers

1. `search_skills`

2. `reload_skills`

3. `get_skill_info`

4. `list_skill_folders`