Project Context MCP Server

A smart context management system for coding projects that uses knowledge graphs and vector databases for intelligent storage and retrieval.

Features

• Zero-friction storage: Just say "add this to context" - no manual categorization needed
• Smart auto-categorization: AI-powered category detection and assignment
• Intelligent relationships: Automatic discovery of connections between contexts
• Semantic search: Find contexts by meaning, not just keywords
• Session continuity: Never lose project context again

Quick Start

Installation with uv (Recommended)

# Install dependencies
uv sync

# Run setup script
uv run python scripts/setup_dev.py

Alternative Installation with pip

pip install -e .
python scripts/setup_dev.py

Setup

1. Copy environment template:

cp .env.example .env

2. Run the MCP server:

uv run python -m src.main
# OR with pip: python -m src.main

MCP Tools

The server exposes 5 simple tools:

1. list_categories() - Show available context categories
2. store_context(content, category) - Store new context with auto-categorization
3. get_context(topic, category) - Retrieve contexts about a topic
4. get_related_contexts(topic) - Find related contexts using knowledge graph
5. delete_context(topic, category) - Remove contexts about a topic

Architecture

• Vector Database: LanceDB for high-performance semantic search with Apache Arrow
• Knowledge Graph: NetworkX for relationship management
• Embeddings: Sentence Transformers (all-MiniLM-L6-v2) for local, fast embeddings
• Auto-categorization: Smart category detection
• Duplicate detection: Prevents redundant contexts
• Local storage: All data stored locally in ./data/

Technology Stack

• LanceDB: Ultra-fast vector database built on Apache Arrow
• NetworkX: Directed graph for context relationships
• Sentence Transformers: Local embedding generation
• Pydantic: Type safety and validation
• Python 3.10+: Modern Python features

Development

# Install development dependencies
uv sync

# Format code
uv run black .
uv run ruff check .

# Type checking
uv run mypy src/

Performance Benefits of LanceDB

• 2-3x faster than ChromaDB for queries
• Lower memory footprint with zero-copy operations
• Built-in analytics capabilities for complex filtering
• Apache Arrow based for maximum performance
• Local-first with no external dependencies

License

MIT