# MCP Knowledge Server - Configuration File
#
# This is the default configuration. Copy this file to config.yaml
# in the project root and customize as needed.
#
# Note: Environment variables take precedence over this file.
# Format: KNOWLEDGE_<SECTION>__<KEY>=value
# Example: KNOWLEDGE_STORAGE__DOCUMENTS_PATH=/custom/path
# Storage Configuration
storage:
# Path where uploaded documents are stored
documents_path: ./data/documents
# Path for ChromaDB vector database
vector_db_path: ./data/chromadb
# Cache directory for downloaded models
model_cache_path: ~/.cache/huggingface
# Embedding Model Configuration
embedding:
# HuggingFace model name for embeddings
# all-MiniLM-L6-v2: Fast, 384 dimensions, good quality
# Other options:
# - all-mpnet-base-v2: Better quality, 768 dimensions, slower
# - paraphrase-multilingual-MiniLM-L12-v2: Multi-language support
model_name: sentence-transformers/all-MiniLM-L6-v2
# Batch size for embedding generation
# Higher = faster but more memory
batch_size: 32
# Device for model inference
# Options: cpu, cuda, mps (for Apple Silicon)
device: cuda
# Text Chunking Configuration
chunking:
# Target chunk size in characters
# Larger chunks = more context, fewer chunks
# Smaller chunks = more precise, more chunks
chunk_size: 500
# Overlap between chunks in characters
# Helps maintain context across chunk boundaries
chunk_overlap: 50
# Chunking strategy
# Options: sentence, paragraph, fixed
# - sentence: Smart splitting on sentence boundaries
# - paragraph: Split on paragraph breaks
# - fixed: Fixed character length
strategy: sentence
# Document Processing Configuration
processing:
# Maximum number of concurrent document processing tasks
max_concurrent_tasks: 3
# OCR confidence threshold (0.0 to 1.0)
# Text below this confidence will trigger warnings
ocr_confidence_threshold: 0.6
# Maximum file size in megabytes
# Files larger than this will be rejected
max_file_size_mb: 100
# Supported document formats
# Automatically detected, listed here for reference
supported_formats:
- pdf
- docx
- pptx
- xlsx
- html
- htm
- jpg
- jpeg
- png
- svg
# MCP Server Configuration
mcp:
# Server host (0.0.0.0 for all interfaces, 127.0.0.1 for local only)
host: 0.0.0.0
# Server port (only used for HTTP transport)
port: 3000
# Transport protocol
# Options: stdio (recommended for MCP), http (for testing)
transport: stdio
# Enable CORS for HTTP transport (development only)
cors_enabled: false
# Allowed origins for CORS
cors_origins:
- http://localhost:3000
- http://127.0.0.1:3000
# Logging Configuration
logging:
# Log level
# Options: DEBUG, INFO, WARNING, ERROR, CRITICAL
level: INFO
# Log file path (optional, logs to stdout if not specified)
# file: ./mcp_server.log
# Log format
# Options: json, text
format: text
# Enable detailed logging for debugging
debug: false
# Search Configuration
search:
# Default number of results to return
default_top_k: 10
# Maximum number of results allowed
max_top_k: 50
# Default minimum relevance score (0.0 to 1.0)
default_min_relevance: 0.0
# Enable search result highlighting
enable_highlighting: true
# Performance Configuration
performance:
# Enable embedding cache for faster repeated queries
cache_embeddings: true
# Cache size limit (number of embeddings)
cache_size: 1000
# Enable lazy loading of models
lazy_load_models: true
# Connection pool size for vector database
db_pool_size: 5
# Security Configuration (for production)
security:
# Enable authentication (requires implementation)
enable_auth: false
# Allowed file extensions (whitelist)
allowed_extensions:
- .pdf
- .docx
- .pptx
- .xlsx
- .html
- .htm
- .jpg
- .jpeg
- .png
- .svg
# Maximum documents per knowledge base
max_documents: 10000
# Enable content scanning (requires implementation)
scan_uploads: false
# Feature Flags
features:
# Enable OCR for scanned documents
enable_ocr: true
# Enable image analysis (requires additional models)
enable_image_analysis: false
# Enable async processing
enable_async: true
# Enable progress tracking
enable_progress: true
# Enable metadata extraction
enable_metadata: true
# Development Configuration
development:
# Enable debug mode (verbose logging, no caching)
debug_mode: false
# Enable hot reload (for development)
hot_reload: false
# Enable performance profiling
profiling: false
# Mock external services
mock_services: false
# Example: Using environment variables
#
# Instead of editing this file, you can set environment variables:
#
# export KNOWLEDGE_STORAGE__DOCUMENTS_PATH=/custom/path
# export KNOWLEDGE_EMBEDDING__BATCH_SIZE=64
# export KNOWLEDGE_LOGGING__LEVEL=DEBUG
# export KNOWLEDGE_SEARCH__DEFAULT_TOP_K=20
#
# Environment variables override settings in this file.
