cortex-mem-mcp

Step 1: Create Configuration File

Create a config.toml file (e.g., ~/.config/cortex-mem/config.toml):

[cortex]
Data directory for storing memories
data_dir = "~/.cortex-data"
[llm]
LLM API configuration
api_base_url = "https://api.openai.com/v1"
api_key = "your-api-key"
model_efficient = "gpt-4o-mini"
temperature = 0.1
max_tokens = 65536
[embedding]
Embedding configuration
api_base_url = "https://api.openai.com/v1"
api_key = "your-embedding-api-key"
model_name = "text-embedding-3-small"
batch_size = 10
timeout_secs = 30
[qdrant]
Vector database configuration
url = "http://localhost:6333"
collection_name = "cortex_memories"
embedding_dim = 1536
timeout_secs = 30

Step 2: Start Qdrant (Vector Database)

# Using Docker
docker run -d -p 6333:6333 qdrant/qdrant
Verify Qdrant is running
curl http://localhost:6333

Step 3: Configure MCP Client

Configure your MCP client (e.g., Claude Desktop, Cursor, etc.) to use cortex-mem-mcp.

#### Claude Desktop

Edit the configuration file:

Add the following configuration:

{
  "mcpServers": {
    "cortex-memory": {
      "command": "cortex-mem-mcp",
      "args": [
        "--config", "/path/to/config.toml",
        "--tenant", "default"
      ],
      "env": {
        "RUST_LOG": "info"
      }
    }
  }
}

If you built from source, use the full path to the binary:

{
  "mcpServers": {
    "cortex-memory": {
      "command": "/path/to/cortex-mem/target/release/cortex-mem-mcp",
      "args": [
        "--config", "/path/to/config.toml",
        "--tenant", "default"
      ]
    }
  }
}

#### Cursor IDE

Add to your Cursor MCP settings:

{
  "mcpServers": {
    "cortex-memory": {
      "command": "cortex-mem-mcp",
      "args": ["--config", "/path/to/config.toml"]
    }
  }
}

Step 4: Restart Your MCP Client

After configuration, restart Claude Desktop or your MCP client to load the new server.

Step 5: Verify Installation

Test the MCP server manually:

# Run with debug logging
RUST_LOG=debug cortex-mem-mcp --config /path/to/config.toml --tenant default

1. Use meaningful thread IDs - Use descriptive names like project-alpha or user-123-support instead of generic IDs

2. Commit periodically - Call commit after significant conversation milestones to ensure memory extraction

3. Start with search - Before storing new information, search to avoid duplication

4. Use tiered access - Start with abstract or search to find relevant memories, then use overview or content for details

5. Scope your searches - Use the scope parameter to limit searches to relevant sessions