ThoughtMCP

# ThoughtMCP v0.6.0 Release Notes **Release Date:** December 8, 2025 ## Overview ThoughtMCP v0.6.0 introduces natural, cognitive-aligned tool names and numerous quality improvements identified during comprehensive testing. ## ⚠️ Breaking Changes ### Tool Renaming All MCP tools have been renamed to use natural, cognitive-aligned names that better reflect their purpose: | Old Name | New Name | Description | | ------------------------ | ----------- | ---------------------------------------- | | `store_memory` | `remember` | Store new memory with embeddings | | `retrieve_memories` | `recall` | Retrieve memories with composite scoring | | `delete_memory` | `forget` | Delete memory (soft/hard) | | `search_memories` | `search` | Full-text and vector search | | `think_parallel` | `ponder` | Parallel stream reasoning | | `analyze_systematically` | `analyze` | Framework-based analysis | | `decompose_problem` | `breakdown` | Problem decomposition | | `analyze_reasoning` | `evaluate` | Reasoning quality analysis | **Note:** The following tools retain their original names: - `think` - Single-mode reasoning - `update_memory` - Update existing memory - `assess_confidence` - Confidence assessment - `detect_bias` - Bias detection - `detect_emotion` - Emotion analysis ### Migration Guide Update your tool calls to use the new names: ```javascript // Before (v0.5.0) await callTool("store_memory", { content: "...", userId: "..." }); await callTool("retrieve_memories", { userId: "...", text: "..." }); await callTool("think_parallel", { problem: "..." }); // After (v0.6.0) await callTool("remember", { content: "...", userId: "..." }); await callTool("recall", { userId: "...", text: "..." }); await callTool("ponder", { problem: "..." }); ``` ## Quality Improvements - **Full-text search**: Fixed NOT operator handling in QueryParser - **Bias detection**: Added text-based bias detection with correction suggestions - **Memory-augmented reasoning**: Cognitive tools now retrieve relevant memories - **Waypoint graph**: Fixed link creation integration - **Problem decomposition**: Improved sub-problem naming quality - **Framework selection**: Better confidence calibration - **Reasoning specificity**: Problem-specific insights instead of generic recommendations - **Content validation**: Added length validation (10-100,000 characters) - **Metadata merge**: Partial updates preserve existing fields - **Evidence extraction**: Automatic extraction from reasoning text ## New Components - `BiasCorrector` - Generates correction suggestions for detected biases - `ContentValidator` - Validates memory content length - `MetadataMerger` - Handles partial metadata updates - `EvidenceExtractor` - Extracts evidence from reasoning text - `MemoryAugmentedReasoning` - Integrates memory retrieval with cognitive tools - `ProblemComplexityAnalyzer` - Scales analysis depth based on complexity --- # ThoughtMCP v0.5.0 Release Notes **Release Date:** December 7, 2025 ## Overview ThoughtMCP v0.5.0 is a complete architectural rebuild delivering a production-ready AI cognitive architecture with human-like memory and reasoning capabilities. This release represents a ground-up reimplementation with PostgreSQL persistence, parallel reasoning streams, and comprehensive metacognitive monitoring. ## Key Features ### 🧠 Hierarchical Memory Decomposition (HMD) - **Five-Sector Embeddings**: Episodic, Semantic, Procedural, Emotional, and Reflective memory types - **Waypoint Graph System**: Sparse graph with 1-3 connections per memory for efficient traversal - **Composite Scoring**: 0.6×similarity + 0.2×salience + 0.1×recency + 0.1×link_weight - **Temporal Decay**: Exponential forgetting with automatic reinforcement on access - **PostgreSQL Persistence**: Production-grade storage with pgvector for vector operations ### ⚡ Performance - **Sub-200ms Retrieval**: p50 <100ms, p95 <200ms, p99 <500ms at 100k memories - **Fast Embedding**: <500ms for all five sectors - **Parallel Reasoning**: <30s total, <10s per stream - **Efficient Operations**: <100ms confidence assessment, <15% bias detection overhead ### 🔀 Parallel Reasoning Streams - **Four Concurrent Streams**: Analytical, Creative, Critical, and Synthetic reasoning - **Real-Time Coordination**: Synchronization at 25%, 50%, 75% completion - **Conflict Preservation**: Maintains diverse perspectives in synthesis - **Low Overhead**: <10% coordination cost ### 🎯 Dynamic Framework Selection - **Eight Frameworks**: Scientific Method, Design Thinking, Systems Thinking, Critical Thinking, Creative Problem Solving, Root Cause Analysis, First Principles, Scenario Planning - **Auto-Selection**: >80% accuracy in choosing optimal framework - **Hybrid Support**: Combines 2-3 frameworks for complex problems - **Adaptive Learning**: Improves selection over time ### 🔬 Metacognitive Monitoring - **Confidence Calibration**: ±10% accuracy between predicted and actual performance - **Bias Detection**: >70% detection rate for 8 bias types - **Emotion Detection**: >75% accuracy using Circumplex model - **Self-Improvement**: 5-10% monthly performance improvement ### 🏗️ Production Hardening - **96%+ Test Coverage**: 3457 tests across unit, integration, e2e, performance, and accuracy - **Zero TypeScript Errors**: Full type safety throughout - **Security Hardening**: Input validation, rate limiting, secrets management - **Monitoring**: Structured logging, metrics collection, health checks ## Quality Metrics | Metric | Target | Achieved | | ------------------ | ------ | -------- | | Statement Coverage | 95%+ | 96.06% | | Branch Coverage | 90%+ | 91.22% | | Function Coverage | 95%+ | 98.94% | | Test Count | - | 3457 | | TypeScript Errors | 0 | 0 | | ESLint Errors | 0 | 0 | ## Accuracy Targets | Capability | Target | Status | | ---------------------- | ------ | ------------------ | | Confidence Calibration | ±10% | ✅ Validated | | Bias Detection | >70% | ✅ Validated | | Emotion Detection | >75% | ✅ Validated | | Framework Selection | >80% | ✅ Validated | | Memory Retrieval | >85% | ✅ Validated (90%) | ## MCP Tools All cognitive capabilities are exposed through MCP tools: ### Memory Operations - `remember` - Store memories with five-sector embeddings - `recall` - Retrieve with composite scoring - `update_memory` - Update memory content and metadata - `forget` - Delete with cascade options - `search` - Full-text and metadata search ### Reasoning Operations - `think` - Human-like reasoning with multiple modes - `analyze` - Framework-based problem solving - `ponder` - Multi-stream parallel reasoning - `breakdown` - Complex problem breakdown ### Metacognitive Operations - `assess_confidence` - Multi-dimensional confidence assessment - `detect_bias` - Real-time bias detection and correction - `detect_emotion` - Circumplex and discrete emotion analysis - `evaluate` - Reasoning quality assessment ## New in v0.5.0 This is the initial production-ready release of ThoughtMCP with the complete cognitive architecture: - **PostgreSQL with pgvector**: Production-grade vector storage - **Local embeddings**: Uses Ollama, E5, or BGE models (zero API costs) - **Complete MCP integration**: Full tool schemas with validation - **Comprehensive documentation**: User guides, API docs, deployment guides ## Getting Started See [docs/user-guide.md](docs/user-guide.md) for installation and setup instructions. ## Documentation - [User Guide](docs/user-guide.md) - Getting started and workflows - [API Reference](docs/api.md) - Complete API documentation - [MCP Tools](docs/mcp-tools.md) - Tool schemas and examples - [Architecture](docs/architecture.md) - System design - [Deployment](docs/deployment.md) - Production deployment guide - [Development](docs/development.md) - Development workflow ## Requirements - **Node.js**: 18.0+ - **PostgreSQL**: 14.0+ with pgvector extension - **Embedding Model**: Ollama, E5, or BGE (local) ## Installation ```bash npm install thoughtmcp ``` Or clone and build: ```bash git clone https://github.com/keyurgolani/ThoughtMcp.git cd ThoughtMcp npm install npm run build ``` ## Acknowledgments Thank you to all contributors who helped make this release possible. ## License MIT License - see [LICENSE](LICENSE) for details.