# Semem MCP Server This MCP (Model Context Protocol) server provides access to Semem core, Ragno knowledge graph, and ZPT APIs for semantic memory management and knowledge processing. **Now with full GraphRAG compatibility!** ## Features ### 🆕 GraphRAG Standard Tools #### Document Management - `store_document` - Store documents with metadata and vector embeddings - `list_documents` - List and filter stored documents with pagination - `delete_documents` - Remove documents from storage (with confirmation) #### Relationship Management - `create_relations` - Create typed relationships between entities - `search_relations` - Query relationships by entity, type, or direction - `delete_relations` - Remove relationships from knowledge graph #### Hybrid Search (Core GraphRAG Feature) - `hybrid_search` - **Combined vector similarity + graph traversal search** - Configurable vector/graph weights - Multi-hop graph traversal - ZPT navigation integration - Comprehensive result scoring #### Graph Traversal & Analytics - `search_nodes` - Discover and filter graph nodes by type or query - `read_graph` - Export graph structure (adjacency, edge list, Cytoscape formats) - `get_knowledge_graph_stats` - Comprehensive analytics and connectivity metrics #### Enhanced Retrieval - `search_documentation` - Advanced semantic document search with filtering - `add_observations` - Enrich entities with contextual observations ### Semem Core API Tools - `semem_store_interaction` - Store new interactions with embeddings and concepts - `semem_retrieve_memories` - Search for relevant memories based on similarity - `semem_generate_response` - Generate responses using memory context - `semem_generate_embedding` - Generate vector embeddings for text - `semem_extract_concepts` - Extract semantic concepts from text ### Ragno Knowledge Graph Tools - `ragno_decompose_corpus` - Decompose text corpus into RDF entities and relationships - `ragno_create_entity` - Create new RDF entities with ontology compliance - `ragno_create_semantic_unit` - Create semantic text units with metadata ### ZPT (Zoom, Pan, Tilt) Navigation Tools - `zpt_select_corpuscles` - Multi-dimensional content selection with 3D navigation - `zpt_chunk_content` - Advanced content chunking with semantic boundaries ### Resources - `semem://status` - System status and service health information - `semem://graph/schema` - RDF graph schema and ontology documentation - `semem://docs/api` - Complete API documentation ## Usage ### Starting the Server ```bash # Start with stdio transport (for use with MCP clients) node mcp/index.js # Test with MCP inspector npx @modelcontextprotocol/inspector node mcp/index.js # Custom ports if defaults are in use CLIENT_PORT=8080 SERVER_PORT=9000 npx @modelcontextprotocol/inspector node mcp/index.js ``` ### Testing ```bash # Run the test script node mcp/test-server.js ``` ### Configuration The server requires: - A valid `config/config.json` file with LLM provider settings - Environment variables for API keys (see main project `.env` file) - Node.js 20.11.0+ ### Example Tool Calls #### 🆕 GraphRAG Examples ##### Store a Document ```json { "name": "store_document", "arguments": { "content": "Artificial intelligence (AI) is intelligence demonstrated by machines...", "metadata": { "title": "Introduction to AI", "author": "AI Research Team", "type": "research", "tags": ["ai", "technology", "research"] } } } ``` ##### Hybrid Search (Core GraphRAG) ```json { "name": "hybrid_search", "arguments": { "query": "machine learning applications", "options": { "maxResults": 15, "vectorWeight": 0.6, "graphWeight": 0.4, "graphDepth": 3, "includeDocuments": true, "includeEntities": true, "includeRelationships": true } } } ``` ##### Create Relationships ```json { "name": "create_relations", "arguments": { "sourceEntity": "artificial_intelligence", "targetEntity": "machine_learning", "relationshipType": "includes", "description": "AI includes machine learning as a subset", "weight": 0.9 } } ``` ##### Read Graph Structure ```json { "name": "read_graph", "arguments": { "rootNodes": ["artificial_intelligence"], "maxDepth": 2, "format": "cytoscape", "includeMetadata": true } } ``` ##### Search Documentation ```json { "name": "search_documentation", "arguments": { "query": "neural networks deep learning", "options": { "maxResults": 10, "sortBy": "relevance", "documentTypes": ["research", "tutorial"], "includeContent": true } } } ``` #### Traditional Semem Examples ##### Store an Interaction ```json { "name": "semem_store_interaction", "arguments": { "prompt": "What is machine learning?", "response": "Machine learning is a subset of artificial intelligence...", "metadata": {"topic": "AI", "difficulty": "beginner"} } } ``` ##### Retrieve Memories ```json { "name": "semem_retrieve_memories", "arguments": { "query": "artificial intelligence", "threshold": 0.7, "limit": 5 } } ``` ##### Decompose Text Corpus ```json { "name": "ragno_decompose_corpus", "arguments": { "textChunks": ["AI is transforming technology.", "Machine learning enables pattern recognition."], "options": {"maxEntities": 50, "extractRelationships": true} } } ``` ##### ZPT Content Navigation ```json { "name": "zpt_select_corpuscles", "arguments": { "zoom": "entity", "tilt": "embedding", "selectionType": "embedding", "criteria": {"query": "machine learning"}, "limit": 20 } } ``` ## Integration ### Claude Desktop Add to your Claude Desktop configuration: ```json { "mcpServers": { "semem": { "command": "node", "args": ["path/to/semem/mcp/index.js"], "env": { "NODE_ENV": "production" } } } } ``` ### Other MCP Clients The server follows the standard MCP protocol and should work with any compatible client using stdio transport. ## Dependencies - `@modelcontextprotocol/sdk` - MCP SDK for server implementation - `zod` - Schema validation - Semem core modules (automatically imported from parent project) ## GraphRAG Compatibility This server provides **full compatibility** with standard GraphRAG MCP patterns while adding unique Semem extensions: ### ✅ Standard GraphRAG Tools Supported - Document storage and retrieval - Entity relationship management - Hybrid vector + graph search - Graph traversal and analytics - Knowledge graph statistics - Entity observations and enrichment ### 🚀 Semem Extensions - **ZPT (Zoom, Pan, Tilt)**: 3D navigation through knowledge space - **Ragno RDF Compliance**: Full semantic web ontology integration - **Multi-Tilt Representations**: Multiple perspectives on the same content - **Semantic Memory Integration**: Persistent conversational memory ### Tool Count: **17 Total** - **9 GraphRAG Standard Tools**: `store_document`, `list_documents`, `delete_documents`, `create_relations`, `search_relations`, `delete_relations`, `hybrid_search`, `search_nodes`, `read_graph`, `get_knowledge_graph_stats`, `search_documentation`, `add_observations` - **5 Semem Core Tools**: Memory management and LLM integration - **3 Ragno Tools**: RDF knowledge graph construction - **2 ZPT Tools**: Multi-dimensional content navigation ## Architecture The MCP server acts as a bridge between the MCP protocol and the Semem APIs, now with full GraphRAG compatibility: ``` MCP Client → MCP Server → GraphRAG Standard APIs → Semem Core APIs → Ragno Knowledge Graph (RDF) → ZPT Navigation/Transform ``` ### Data Flow 1. **Documents** → Vector embeddings + RDF entities + Graph relationships 2. **Hybrid Search** → Vector similarity + Graph traversal + ZPT navigation 3. **Results** → Scored and ranked using multiple strategies 4. **Storage** → Persistent memory + RDF graph + Relationship index Each tool call is validated, executed against the appropriate API layer, and results are returned in MCP-compliant format with comprehensive error handling and demo fallbacks.