# MCP Skills Configuration Example
# Copy this file to ~/.mcp-skillkit/config.yaml and customize as needed
# Hybrid Search Configuration
# Controls the weighting between vector similarity search and knowledge graph relationships
hybrid_search:
# Option 1: Use a preset (recommended)
# Available presets:
# - current: Default optimized preset (70% vector, 30% graph)
# - semantic_focused: Best for natural language queries (90% vector, 10% graph)
# - graph_focused: Best for finding related skills (30% vector, 70% graph)
# - balanced: Equal weighting for general purpose (50% vector, 50% graph)
preset: current
# Option 2: Specify custom weights (must sum to 1.0)
# Uncomment and adjust these lines to use custom weights instead of a preset
# vector_weight: 0.7 # Weight for vector similarity search (0.0-1.0)
# graph_weight: 0.3 # Weight for knowledge graph relationships (0.0-1.0)
# Vector Store Configuration
vector_store:
backend: chromadb # chromadb, qdrant, or faiss
embedding_model: all-MiniLM-L6-v2 # Sentence transformer model
collection_name: skills_v1 # ChromaDB collection name
# Knowledge Graph Configuration
knowledge_graph:
backend: networkx # networkx or neo4j
# MCP Server Configuration
server:
transport: stdio # stdio, http, or sse
port: 8000 # Port for HTTP transport
log_level: info # debug, info, warning, or error
max_loaded_skills: 50 # Maximum skills to keep in memory
# Toolchain Detection
toolchain_cache_duration: 3600 # Cache duration in seconds
auto_recommend: true # Automatically recommend skills on toolchain detection
# Repository Configuration (optional)
# repositories:
# - url: https://github.com/anthropics/mcp-skillkit-core
# priority: 90
# auto_update: true
# - url: https://github.com/your-org/custom-skills
# priority: 80
# auto_update: true
# Use Cases by Preset:
#
# current (0.7 vector, 0.3 graph) - DEFAULT
# Best for: General-purpose skill discovery
# Use when: You want balanced results with slight emphasis on semantic matching
# Example: "python testing frameworks" → finds pytest, unittest with related testing skills
#
# semantic_focused (0.9 vector, 0.1 graph)
# Best for: Natural language queries, fuzzy matching, concept-based search
# Use when: You have vague requirements or want semantic similarity
# Example: "help me debug async code" → emphasizes semantic understanding over relationships
#
# graph_focused (0.3 vector, 0.7 graph)
# Best for: Discovering related skills, exploring dependencies, finding skill clusters
# Use when: You want to explore what skills work together or depend on each other
# Example: Starting from "pytest" → discovers pytest-fixtures, pytest-mock, test-coverage
#
# balanced (0.5 vector, 0.5 graph)
# Best for: General purpose when you want equal emphasis
# Use when: You're unsure which approach is better for your query
# Example: "javascript testing" → equal weight to semantic match and relationships
# CLI Usage Examples:
#
# Use config file preset:
# mcp-skillkit search "python testing"
#
# Override with CLI flag:
# mcp-skillkit search "python testing" --search-mode semantic_focused
# mcp-skillkit recommend --search-mode graph_focused
#
# Available CLI flags:
# --search-mode semantic_focused # 90% vector, 10% graph
# --search-mode graph_focused # 30% vector, 70% graph
# --search-mode balanced # 50% vector, 50% graph
# --search-mode current # 70% vector, 30% graph (default)
