Zero-Vector MCP acts like a high-powered digital brain for AI assistants, giving them the ability to remember past conversations and user preferences long after a session has ended. Instead of an AI "forgetting" everything once a chat window is closed, this tool provides a persistent filing cabinet where it can store and retrieve information. It allows developers to create distinct AI personas with their own unique memories, making interactions feel more personalized and contextually aware over time. For developers building with AI tools like Cline, this system offers a production-ready bridge between raw language models and long-term data storage. By utilizing the Model Context Protocol, it provides a suite of 13 specialized tools that allow an AI to autonomously manage its own memory. It can search for conceptually related information through semantic search, assign importance scores to specific facts, and even prune old or irrelevant data to keep its context window clean and efficient. Under the hood, the system is engineered for high-speed performance, boasting sub-50ms query times and a massive capacity for over 349,000 vectors. The architecture combines a specialized vector database for similarity matching with a SQLite backend for structured metadata, ensuring that data retrieval is both lightning-fast and highly organized. With built-in security features like role-based API key authentication and multi-tier rate limiting, it offers a scalable and secure foundation for creating sophisticated AI agents that require instant access to a vast library of stored knowledge.
Category: AI Memory & Context
Tags: persistence, rag, semantic search, sqlite, vector-database
bash git clone https://github.com/MushroomFleet/zero-vector-MCP.git cd zero-vector-MCP Step 2: Set up the Zero-Vector server bash cd zero-vector/server npm install npm run setup:database npm run generate:api-key # Note the generated key for the MCP config cp env.example .env # Add your OpenAI API key to this file npm start Step 3: Set up the MCP server bash # Open a new terminal cd MCP npm install cp env.example .env # Edit .env with your Zero-Vector server URL and the API key generated in Step 2 npm start ---json { "mcpServers": { "zero-vector": { "command": "node", "args": ["C:/path/to/zero-vector-MCP/MCP/src/index.js"], "env": { "ZERO_VECTOR_BASE_URL": "http://localhost:3000", "ZERO_VECTOR_API_KEY": "your_api_key_here" } } } } Environment Variables for MCP Server: * ZERO_VECTOR_BASE_URL: The URL where the core server is running (default: http://localhost:3000). * ZERO_VECTOR_API_KEY: The key generated during the server setup. * LOG_LEVEL: Logging verbosity (e.g., info). ---create_persona: Configure a new AI persona with custom behavior settings. * list_personas: List all existing personas. * get_persona: Retrieve details for a specific persona. * update_persona: Modify persona settings. * delete_persona: Remove a persona. Memory & Conversation Tools: * add_memory: Store context-aware memory with importance scoring. * search_persona_memories: Find relevant memories using semantic vector similarity. * add_conversation: Log conversation history. * get_conversation_history: Retrieve complete conversation tracking. * cleanup_persona_memories: Automated cleanup of low-importance or old memories. Utility Tools: * get_system_health: Check database and server status. * get_persona_stats: View statistics for a specific persona. * test_connection: Verify the link between the MCP server and the vector database. ---add_memory with an importance score of 0.9. In a new session, when the developer says "Write a test for this utility," the AI calls search_persona_memories for "testing framework" and automatically generates Vitest code instead of Jest.search_persona_memories with the query "authentication flow." It retrieves the specific vector-stored documentation and provides an accurate answer without having the entire manual in its active context.create_persona, developers can store key narrative events as memories. The "Importance Scoring" feature ensures that critical plot points stay in memory while trivial chatter can be removed via cleanup_persona_memories. Example: A player betrays an NPC. The AI adds a memory: "Player lied about the stolen artifact" with a type of "relationship_event" and high importance. Fifty chat turns later, the AI searches the NPC's memory for "player trust" and accurately reflects the character's lingering resentment…Part of MCP Servers