home / mcp / titan mcp server

Titan MCP Server

Provides a memory-enabled MCP server that stores, recalls, and learns from user interactions across sessions.

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "henryhawke-mcp-titan": {
      "command": "npx",
      "args": [
        "@henryhawke/mcp-titan"
      ]
    }
  }
}

You can run and connect to a Titan-based MCP server that remembers context across conversations, enabling three-tier memory, continuous learning, and sequence-aware recall to deliver more personalized interactions over time.

How to use

You will install the MCP Titan server, start it, and then connect any MCP client (Cursor, Claude Desktop, or your own interface) to store, recall, and learn from memories. Use the server to persist user context, recall relevant past details during chats, and progressively improve responses through online learning. Start the server, initialize its memory and learner, and then issue store_memory and recall requests through your MCP client to keep track of what the user discussed, what should be remembered long term, and what should be kept in archive storage.

How to install

Prerequisites: Node.js 22 or newer, and npm, bun, or another compatible package manager.

1) Install the MCP Titan package using your preferred package manager.

# Install via npm
npm install @henryhawke/mcp-titan

# Or via bun
bun add @henryhawke/mcp-titan

2) Start the HOPE/MCP Titan server from the command line.

npx @henryhawke/mcp-titan

3) Verify the server creates a memory storage directory and begins listening for MCP tool calls.

# The server will initialize storage at ~/.hope_memory and start accepting MCP commands.

Available tools

init_model

Initialize or reconfigure the memory system and its three-tier storage.

bootstrap_memory

Preload memory with initial context from text or URL.

get_memory_state

Inspect current memory contents across short-term, long-term, and archive.

memory_stats

Obtain a statistical summary of memory usage and distribution.

prune_memory

Clean up low-value memories to prevent memory bloat.

forward_pass

Query memory and get a prediction or recall based on current context.

train_step

Explicitly learn from a given input-output example pair.

reset_gradients

Clear training state and gradients during online learning.

save_checkpoint

Save the current memory state to a file.

load_checkpoint

Restore memory state from a saved checkpoint.

export_checkpoint

Export the current checkpoint to a portable file.

import_checkpoint

Import a checkpoint from a file to restore state.

init_learner

Start the online learning service to update memory from interactions.

pause_learner

Pause the background learning process without losing state.

resume_learner

Resume the background learning process.

get_learner_stats

Monitor learning metrics and progress of the online learner.

add_training_sample

Provide input-target pairs to train the memory system in real time.

get_token_flow_metrics

Analyze sequence patterns and token flow through memory.

get_hierarchical_metrics

Monitor distribution across short-term, long-term, and archive tiers.

health_check

Assess system health and performance.