mcp-skillkit

Dynamic RAG-powered skills for code assistants via Model Context Protocol (MCP)

mcp-skillkit is a standalone Python application that provides intelligent, context-aware skills to code assistants through hybrid RAG (vector + knowledge graph). Unlike static skills that load at startup, mcp-skillkit enables runtime skill discovery, automatic recommendations based on your project's toolchain, and dynamic loading optimized for your workflow.

Key Features

• 🚀 Zero Config: mcp-skillkit setup handles everything automatically
• 🧠 Intelligent: Auto-detects your project's toolchain (Python, TypeScript, Rust, Go, etc.)
• 🔍 Dynamic Discovery: Vector similarity + knowledge graph for better skill finding
• 📦 Multi-Source: Pulls skills from multiple git repositories
• ⚡ On-Demand Loading: Skills loaded when needed, not all at startup
• 🔌 MCP Native: First-class Model Context Protocol integration

Installation

With pipx (Recommended)

pipx is the recommended way to install Python CLI applications:

pipx install mcp-skillkit

With pip

If you prefer pip (not recommended for CLI tools):

pip install mcp-skillkit

From Source

git clone https://github.com/bobmatnyc/mcp-skillkit.git
cd mcp-skillkit
pip install -e .

Local Development (Without Installation)

For development, you can run mcp-skillkit directly from source without installing:

# Use the development script
./mcp-skillkit-dev --help
./mcp-skillkit-dev search "python testing"
./mcp-skillkit-dev setup --auto

The mcp-skillkit-dev script:

• Runs the package from source code (not installed version)
• Uses local virtual environment if available
• Sets up PYTHONPATH automatically
• Passes all arguments through to the CLI

This is useful for:

• Testing changes without reinstalling
• Developing new features
• Debugging with source code
• Contributing to the project

Note: For production use, install the package normally with pip install -e . or pip install mcp-skillkit.

First-Run Requirements

Important: On first run, mcp-skillkit will automatically download a ~90MB sentence-transformer model (all-MiniLM-L6-v2) for semantic search. This happens during the initial mcp-skillkit setup or when you first run any command that requires indexing.

Requirements:

• ✅ Active internet connection
• ✅ ~100MB free disk space
• ✅ 2-5 minutes for initial download (depending on connection speed)

Model Caching:

• Models are cached in ~/.cache/huggingface/ for future use
• Subsequent runs use the cached model (no download required)
• The cache persists across mcp-skillkit updates

Quick Start

1. Setup

Run the interactive setup wizard to configure mcp-skillkit for your project:

mcp-skillkit setup

Note: The first run will download the embedding model (~90MB) before proceeding with setup. Allow 2-5 minutes for this initial download. Subsequent runs will be much faster.

This will:

• Download embedding model (first run only)
• Detect your project's toolchain
• Clone relevant skill repositories
• Build vector + knowledge graph indices
• Configure MCP server integration
• Validate the setup

2. Start the MCP Server

mcp-skillkit serve

The server will start and expose skills to your code assistant via MCP protocol.

3. Use with Claude Code

Skills are automatically available in Claude Code. Try:

• "What testing skills are available for Python?"
• "Show me debugging skills"
• "Recommend skills for my project"

Project Structure

~/.mcp-skillkit/
├── config.yaml              # User configuration
├── repos/                   # Cloned skill repositories
│   ├── anthropics/skills/
│   ├── obra/superpowers/
│   └── custom-repo/
├── indices/                 # Vector + KG indices
│   ├── vector_store/
│   └── knowledge_graph/
└── metadata.db             # SQLite metadata

Architecture

mcp-skillkit uses a hybrid RAG approach combining:

Vector Store (ChromaDB):

• Fast semantic search over skill descriptions
• Embeddings generated with sentence-transformers
• Persistent local storage with minimal configuration

Knowledge Graph (NetworkX):

• Skill relationships and dependencies
• Category and toolchain associations
• Related skill discovery

Toolchain Detection:

• Automatic detection of programming languages
• Framework and build tool identification
• Intelligent skill recommendations

Configuration

Global Configuration (~/.mcp-skillkit/config.yaml)

# Hybrid Search Configuration
# Controls weighting between vector similarity and knowledge graph relationships
hybrid_search:
  # Option 1: Use a preset (recommended)
  preset: current  # current, semantic_focused, graph_focused, or balanced

  # Option 2: Specify custom weights (must sum to 1.0)
  # vector_weight: 0.7  # Weight for vector similarity (0.0-1.0)
  # graph_weight: 0.3   # Weight for knowledge graph (0.0-1.0)

repositories:
  - url: https://github.com/anthropics/skills.git
    priority: 100
    auto_update: true

vector_store:
  backend: chromadb
  embedding_model: all-MiniLM-L6-v2

server:
  transport: stdio
  log_level: info

Hybrid Search Modes

The hybrid search system combines vector similarity (semantic search) with knowledge graph relationships (dependency traversal) to find relevant skills. You can tune the weighting to optimize for different use cases:

Available Presets:

Preset	Vector	Graph	Best For	Use Case
current	70%	30%	General purpose (default)	Balanced skill discovery with slight semantic emphasis
semantic_focused	90%	10%	Natural language queries	"help me debug async code" → emphasizes semantic understanding
graph_focused	30%	70%	Related skill discovery	Starting from "pytest" → discovers pytest-fixtures, pytest-mock
balanced	50%	50%	Equal weighting	General purpose when unsure which approach is better

When to use each mode:

• current (default): Best for most users. Proven through testing to work well for typical skill discovery patterns.
• semantic_focused: Use when you have vague requirements or want fuzzy semantic matching. Good for concept-based searches like "help me with error handling" or "testing strategies".
• graph_focused: Use when you want to explore skill ecosystems and dependencies. Perfect for "what else works with X?" queries.
• balanced: Use when you want equal emphasis on both approaches, or as a starting point for experimentation.

Configuration Examples:

# Use preset (recommended)
hybrid_search:
  preset: current

# OR specify custom weights
hybrid_search:
  vector_weight: 0.8
  graph_weight: 0.2

CLI Override:

You can override the config file setting using the --search-mode flag:

# Use semantic-focused mode for this search
mcp-skillkit search "python testing" --search-mode semantic_focused

# Use graph-focused mode for recommendations
mcp-skillkit recommend --search-mode graph_focused

# Available modes: semantic_focused, graph_focused, balanced, current

Project Configuration (.mcp-skillkit.yaml)

project:
  name: my-project
  toolchain:
    primary: Python
    frameworks: [Flask, SQLAlchemy]

auto_load:
  - systematic-debugging
  - test-driven-development

CLI Commands

# Setup and Configuration
mcp-skillkit setup                    # Interactive setup wizard
mcp-skillkit config                   # Show configuration

# Server
mcp-skillkit serve                    # Start MCP server (stdio)
mcp-skillkit serve --http             # Start HTTP server
mcp-skillkit serve --dev              # Development mode (auto-reload)

# Skills Management
mcp-skillkit search "testing"         # Search skills
mcp-skillkit list                     # List all skills
mcp-skillkit info pytest-skill        # Show skill details
mcp-skillkit recommend                # Get recommendations

# Repositories
mcp-skillkit repo add <url>           # Add repository
mcp-skillkit repo list                # List repositories
mcp-skillkit repo update              # Update all repositories

# Indexing
mcp-skillkit index                    # Rebuild indices
mcp-skillkit index --incremental      # Index only new skills

# Utilities
mcp-skillkit health                   # Health check
mcp-skillkit stats                    # Usage statistics

Shell Completions

Enable tab completion for the mcp-skillkit command to speed up your workflow:

Quick Install

Bash (requires Bash 4.4+):

eval "$(_MCP_SKILLS_COMPLETE=bash_source mcp-skillkit)" >> ~/.bashrc
source ~/.bashrc

Zsh (macOS default):

eval "$(_MCP_SKILLS_COMPLETE=zsh_source mcp-skillkit)" >> ~/.zshrc
source ~/.zshrc

Fish:

echo 'eval (env _MCP_SKILLS_COMPLETE=fish_source mcp-skillkit)' >> ~/.config/fish/config.fish
source ~/.config/fish/config.fish

Features

• ✅ Complete all commands and subcommands
• ✅ Complete option flags (--help, --limit, etc.)
• ✅ Works with mcp-skillkit, mcp-skillkit repo, and all other commands

Verification

Test completions are working:

mcp-skillkit <TAB>        # Shows: config health index info list mcp recommend repo search setup stats
mcp-skillkit repo <TAB>   # Shows: add list update
mcp-skillkit search --<TAB>  # Shows: --category --help --limit

Documentation

For detailed installation instructions, troubleshooting, and advanced usage, see docs/SHELL_COMPLETIONS.md.

MCP Tools

mcp-skillkit exposes these tools to code assistants:

• search_skills: Natural language skill search
• get_skill: Load full skill instructions by ID
• recommend_skills: Get recommendations for current project
• list_categories: List all skill categories
• update_repositories: Pull latest skills from git

Development

Requirements

• Python 3.11+
• Git

Setup Development Environment

git clone https://github.com/bobmatnyc/mcp-skillkit.git
cd mcp-skillkit
pip install -e ".[dev]"

Running from Source (Development Mode)

Use the ./mcp-skillkit-dev script to run commands directly from source without installation:

# Run any CLI command
./mcp-skillkit-dev --version
./mcp-skillkit-dev search "debugging"
./mcp-skillkit-dev serve --dev

# All arguments pass through
./mcp-skillkit-dev info systematic-debugging

How it works:

1. Sets PYTHONPATH to include src/ directory
2. Activates local .venv if present
3. Runs python -m mcp_skills.cli.main with all arguments

When to use:

• ✅ Rapid iteration during development
• ✅ Testing changes without reinstalling
• ✅ Debugging with source code modifications
• ❌ Production deployments (use pip install instead)

Installed vs. Source:

# Installed version (from pip install -e .)
mcp-skillkit search "testing"

# Source version (no installation required)
./mcp-skillkit-dev search "testing"

Run Tests

make quality

Performance Benchmarks

mcp-skillkit includes comprehensive performance benchmarks to track and prevent regressions:

# Run all benchmarks (includes slow tests)
make benchmark

# Run fast benchmarks only (skip 10k skill tests)
make benchmark-fast

# Compare current performance with baseline
make benchmark-compare

Benchmark Categories:

• Indexing Performance: Measure time to index 100, 1000, and 10000 skills
• Search Performance: Track query latency (p50, p95, p99) for vector and hybrid search
• Database Performance: Benchmark SQLite operations (lookup, query, batch insert)
• Memory Usage: Monitor memory consumption during large-scale operations

Baseline Thresholds:

• Index 100 skills: < 10 seconds
• Index 1000 skills: < 100 seconds
• Search query (p50): < 100ms
• Search query (p95): < 500ms
• SQLite lookup by ID: < 1ms

Benchmark Results:

• Results are saved to .benchmarks/ directory (git-ignored)
• Use make benchmark-compare to detect performance regressions
• CI/CD can be configured to fail on significant performance degradation

Example Output:

-------------------------- benchmark: 15 tests --------------------------
Name (time in ms)                    Min      Max     Mean   StdDev
---------------------------------------------------------------------
test_vector_search_latency_100      45.2     52.1    47.8     2.1
test_lookup_by_id_single             0.3      0.8     0.4     0.1
test_hybrid_search_end_to_end       89.5    105.2    94.3     5.2
---------------------------------------------------------------------

Linting and Formatting

make lint-fix

Security Scanning

mcp-skillkit includes comprehensive security scanning to identify vulnerabilities in dependencies and code:

Automated Security (Dependabot + GitHub Actions)

Dependabot automatically:

• Scans dependencies weekly for vulnerabilities
• Creates pull requests for security updates
• Groups minor/patch updates for easier review

GitHub Actions runs security scans on every push:

• Safety: Python dependency vulnerability scanner
• pip-audit: PyPI package vulnerability auditor
• Bandit: Python code security linter
• detect-secrets: Secret detection scanner

Manual Security Scanning

# Basic security scan (Safety + pip-audit)
make security-check

# Comprehensive security audit with reports
make security-check-full

# Install security scanning tools
make security-install

# Pre-publish with security checks
make pre-publish

Security Reports

After running make security-check-full, reports are saved to .security-reports/:

• safety-report.json - Dependency vulnerabilities
• pip-audit-report.json - Package vulnerabilities
• bandit-report.json - Code security issues

Security Policy

For vulnerability reporting and security best practices, see .github/SECURITY.md.

Key security features:

• Automated dependency scanning (Dependabot)
• Weekly security scans (GitHub Actions)
• Pre-publish security gate
• Secret detection (detect-secrets)
• Code security linting (Bandit)

Documentation

Architecture

See docs/architecture/README.md for detailed architecture design.

Skills Collections

See docs/skills/RESOURCES.md for a comprehensive index of skill repositories compatible with mcp-skillkit, including:

• Official Anthropic skills
• Community collections (obra/superpowers, claude-mpm-skills, etc.)
• Toolchain-specific skills (Python, TypeScript, Rust, Go, Java)
• Operations & DevOps skills
• MCP servers that provide skill-like capabilities

Troubleshooting

Model Download Issues

If you encounter problems downloading the embedding model on first run:

1. Check Internet Connection

The model is downloaded from HuggingFace Hub. Verify you can reach:

curl -I https://huggingface.co

2. Manual Model Download

Pre-download the model manually if automatic download fails:

python -c "from sentence_transformers import SentenceTransformer; SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')"

This downloads the model to ~/.cache/huggingface/ and verifies it works.

3. Proxy Configuration

If behind a corporate proxy, configure environment variables:

export HTTP_PROXY=http://proxy.example.com:8080
export HTTPS_PROXY=http://proxy.example.com:8080
export HF_ENDPOINT=https://huggingface.co  # Or your mirror

4. Offline/Air-Gapped Installation

For environments without internet access:

On a machine with internet:

1. Download the model:

   python -c "from sentence_transformers import SentenceTransformer; SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')"
   

2. Package the model cache:

   cd ~/.cache/huggingface
   tar -czf sentence-transformers-model.tar.gz hub/
   

On the air-gapped machine:

1. Transfer sentence-transformers-model.tar.gz to the target machine

2. Extract to the HuggingFace cache directory:

   mkdir -p ~/.cache/huggingface
   cd ~/.cache/huggingface
   tar -xzf /path/to/sentence-transformers-model.tar.gz
   

3. Install mcp-skillkit (transfer wheel if needed):

   pip install mcp-skillkit  # Or install from wheel
   

4. Verify the setup:

   mcp-skillkit health
   

5. Custom Cache Location

If you need to use a different cache directory:

export HF_HOME=/custom/path/to/cache
export TRANSFORMERS_CACHE=/custom/path/to/cache
mcp-skillkit setup

6. Disk Space Issues

Check available space in the cache directory:

df -h ~/.cache/huggingface

The model requires ~90MB, but allow ~100MB for temporary files during download.

7. Permission Issues

Ensure the cache directory is writable:

mkdir -p ~/.cache/huggingface
chmod 755 ~/.cache/huggingface

Common Issues

"Connection timeout" during model download

• Check internet connection and firewall settings
• Try manual download (see step 2 above)
• Configure proxy if behind corporate network (see step 3 above)

"No space left on device"

• Check disk space: df -h ~/.cache
• Clear old HuggingFace cache: rm -rf ~/.cache/huggingface/*
• Use custom cache location (see step 5 above)

"Permission denied" on cache directory

• Fix permissions: chmod 755 ~/.cache/huggingface
• Or use custom cache location with proper permissions

Slow initial setup

• First run downloads ~90MB and builds indices
• Expected time: 2-10 minutes depending on connection speed and number of skills
• Subsequent runs use cached model and are much faster

Getting Help

If you encounter issues not covered here:

1. Check GitHub Issues
2. Review logs: ~/.mcp-skillkit/logs/
3. Run health check: mcp-skillkit health
4. Open a new issue with:
   • Error message and stack trace
   • Output of mcp-skillkit --version
   • Operating system and Python version
   • Steps to reproduce

Contributing

Contributions welcome! Please read our contributing guidelines first.

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Run make quality to ensure tests pass
5. Submit a pull request

License

MIT License - see LICENSE for details.

Acknowledgments

• Built on the Model Context Protocol
• Inspired by Claude Skills
• Uses ChromaDB for vector search
• Embeddings via sentence-transformers

Links

• PyPI Package: mcp-skillkit on PyPI
• Documentation: GitHub Wiki
• Issue Tracker: GitHub Issues
• MCP Registry: MCP Servers
• Publishing Guide: docs/publishing.md

---

Status: ✅ v0.1.0 - Production Ready | Test Coverage: 85-96% | Tests: 48 passing