mcp-titan

mcp-titan

54
knowledge_and_memory

The Titan Memory MCP Server provides a neural memory system that supports learning and predicting sequences for LLMs, with emphasis on maintaining memory state. It offers integration with Cursor and Claude, featuring efficient memory management and transformer-based architecture.

Titan Memory MCP Server

A neural memory system for LLMs that can learn and predict sequences while maintaining state through a memory vector. This MCP (Model Context Protocol) server provides tools for maintaining memory state across interactions, particularly with Claude 3.7 Sonnet and other LLMs.

Features

  • Perfect for Cursor with automatic MCP yolo mode
  • Transformer-based neural memory architecture
  • Efficient memory management including cleanup
  • MCP integration for Cursor and other clients
  • Text encoding and memory persistence

Available Tools

  • help: Get help about available tools
  • init_model: Initialize model with custom configuration
  • forward_pass: Perform forward pass for predictions
  • train_step: Execute a training step
  • get_memory_state: Retrieve current memory state
  • manifold_step: Update memory directionally
  • prune_memory: Remove less relevant memories
  • save_checkpoint: Save memory state
  • load_checkpoint: Load memory state
  • reset_gradients: Reset gradients

Memory Management

Includes automatic cleanup, encryption, validation, and error recovery to ensure smooth tensor operations.

Architecture

Composed of a modular setup including main server class, neural memory model, vector processing, and tensor management systems.

Contributing

Contributions via Pull Requests are encouraged.

License

Licensed under the MIT License.