# Setting up MCP Titan Memory Server in Cursor
## Quick Setup
1. **Ensure the project is built:**
```bash
cd /Users/henrymayo/Desktop/mcp-titan
npm install
npm run build
```
2. **Test the server manually:**
```bash
echo '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"help","arguments":{}}}' | node index.js
```
3. **Configure Cursor MCP:**
- Open Cursor Settings
- Navigate to "MCP Servers" or "Extensions" > "MCP"
- Add a new server configuration:
```json
{
"mcpServers": {
"titan-memory": {
"command": "node",
"args": ["index.js"],
"cwd": "/Users/henrymayo/Desktop/mcp-titan"
}
}
}
```
4. **Restart Cursor** to load the MCP server
## Available Tools
The Titan Memory MCP server provides these tools:
- **`help`** - Get help about available tools
- **`init_model`** - Initialize the Titan Memory model with custom configuration
- **`forward_pass`** - Perform a forward pass through the model
- **`train_step`** - Execute a training step to update the model
- **`get_memory_state`** - Get current memory state and statistics
- **`manifold_step`** - Update memory along a manifold direction
- **`prune_memory`** - Remove less relevant memories to free up space
- **`save_checkpoint`** - Save memory state to a file
- **`load_checkpoint`** - Load memory state from a file
- **`reset_gradients`** - Reset accumulated gradients
## Usage Example
1. **Initialize the model first:**
```
Use the init_model tool with parameters like:
{
"inputDim": 768,
"memorySlots": 1000,
"transformerLayers": 4
}
```
2. **Process text with forward_pass:**
```
Use forward_pass with text input:
{
"x": "Your text input here"
}
```
3. **Train on sequences:**
```
Use train_step with:
{
"x_t": "Current text",
"x_next": "Next text in sequence"
}
```
## Memory Rule Integration
Add this rule to your Cursor rules to ensure proper usage:
```
Rule Name: mcp-titan-memory
Description: When using MCP Titan memory tools, always initialize the model first with init_model before using other tools. Use appropriate input dimensions (768 recommended) and memory slots based on your use case.
```
## Troubleshooting
- **"Method not found" errors:** Ensure you're using the correct MCP protocol format
- **Shape mismatch errors:** Initialize the model with `init_model` first with appropriate `inputDim`
- **Server not starting:** Check that Node.js is installed and `npm run build` completed successfully
- **Permission errors:** Ensure the working directory is writable for checkpoint files