v1.0.0 - First Stable Release

🎉 First Stable Release!

Highlights

• ⚡ Ultra-fast semantic tool filtering - <10ms filtering latency for 1000+ tools
• 🎯 Local & API embeddings - Choose between local embeddings (fast & private) or API embeddings (highest accuracy)
• 📊 Performance optimizations - 6-8x faster dot product, smart top-K selection, true LRU cache
• 🔄 Flexible input - Accepts chat completion messages or raw strings
• 🎛️ Highly configurable - Tune scoring thresholds, top-k, always-include tools

Features

Core Functionality

• Local embeddings support using @xenova/transformers
• API embeddings support (OpenAI, Voyage, Cohere)
• Semantic tool filtering with cosine similarity
• Context caching for performance optimization
• Support for chat messages and string inputs
• Configurable filtering options (topK, minScore, alwaysInclude, exclude)
• Performance metrics and debug logging
• Comprehensive documentation and examples
• includeServerDescription option for better high-level intent queries

Performance Optimizations

• Loop-Unrolled Dot Product: 6-8x faster vector similarity computation
• Smart Top-K Selection: Hybrid algorithm optimized for different workloads
• True LRU Cache: 20-40% better cache hit rates
• In-Place Vector Normalization: ~30% reduction in memory allocations
• Set-Based Exclusion: O(1) exclusion checking
• Overall: 50-80% faster filtering for typical workloads

Installation

npm install @portkey-ai/mcp-tool-filter

Quick Start

import { MCPToolFilter } from '@portkey-ai/mcp-tool-filter';

// Initialize with local embeddings (recommended)
const filter = new MCPToolFilter({
  embedding: { provider: 'local' }
});

await filter.initialize(mcpServers);

// Filter tools based on context
const result = await filter.filter("Search my emails for the Q4 budget");
console.log(result.tools); // Top 20 most relevant tools

Documentation

See the README for comprehensive documentation, examples, and benchmarks.

What's Changed

See CHANGELOG.md for details.