Technical Specifications: LocalSearch MCP Server
System Requirements:
- Python version: 3.10 or higher
- Memory: Minimum 4GB RAM, recommended 8GB RAM
- Storage: 10GB free space for Wikipedia index
- Operating System: Linux, macOS, Windows
Performance Benchmarks:
Index Build Time:
- 10,000 documents: approximately 2 minutes
- 100,000 documents: approximately 15 minutes
- 1,000,000 documents: approximately 2 hours
Search Performance:
- Average query latency: 50-200ms
- Throughput: 100-500 queries per second
- Index size: ~5GB for 1M documents
Supported File Formats:
- Markdown (.md)
- Plain text (.txt)
- Future support planned: PDF, DOCX, HTML
Search Algorithms:
- BM25: Keyword-based ranking
- Vector embeddings: Semantic similarity using sentence transformers
- Hybrid fusion: Reciprocal Rank Fusion (RRF) combining both approaches
Configuration Options:
- WIKI_SUBSET_SIZE: Number of Wikipedia articles to index (default: 1000000)
- LOCAL_DOCS_PATH: Path to local documents directory
- CHUNK_SIZE: Text chunk size for indexing (default: 512 tokens)
- TOP_K: Number of results to return (default: 5)
API Endpoints:
- search_wikipedia: Search only Wikipedia
- search_local: Search only local files
- search: Search all sources (multi-source search)
Quality Metrics:
- Precision@5: 0.85
- Recall@10: 0.78
- MRR (Mean Reciprocal Rank): 0.82
The system is designed for offline operation with no external API dependencies.
