cortex-memory-mcp

Persistent memory for AI agents. An MCP server that gives any AI agent (Claude, Kiro, Cursor, custom) long-term memory across conversations using semantic search.

What it does

Your AI agent forgets everything between conversations. Cortex fixes that. Memories are stored in the cloud, semantically indexed, and instantly retrievable — so your agent can remember preferences, decisions, project context, and anything else that matters.

Tools provided:

| Tool | Description | |------|-------------| | save_memory | Store facts, preferences, decisions, or insights | | search_memory | Semantic search — find memories by meaning, not keywords | | update_memory | Revise an existing memory when things change | | delete_memory | Remove outdated or incorrect memories |

Memories are automatically enriched with semantic vectors for intelligent retrieval. Each API key gets its own isolated memory space.

Quick start

1. Run the setup command

npx cortex-memory-init

This will:

• Open the dashboard to create your API key
• Detect installed editors (Claude Desktop, Claude Code, Kiro, Cursor)
• Write the MCP config for you

Already have a key? Pass it directly:

npx cortex-memory-init ctx_your_key_here

2. Manual setup

If you prefer to configure manually, add to your MCP config:

{
  "mcpServers": {
    "cortex-memory": {
      "command": "npx",
      "args": ["-y", "cortex-memory-mcp"],
      "env": {
        "CORTEX_API_KEY": "ctx_your_key_here"
      }
    }
  }
}

That's it. Your agent now has persistent memory.

How it works

When your agent encounters important information — a user preference, a technical decision, project context — it saves it to Cortex. On subsequent conversations, the agent searches memory for relevant context and picks up where it left off.

Agent → save_memory("User prefers TypeScript over JavaScript")
         ...next conversation...
Agent → search_memory("language preferences")
       → "User prefers TypeScript over JavaScript"

Environment variables

| Variable | Required | Description | |----------|----------|-------------| | CORTEX_API_KEY | Yes | Your API key from the dashboard | | CORTEX_API_URL | No | Custom API endpoint (defaults to hosted service) |

Features

• Semantic search — find memories by meaning, not exact keywords
• User isolation — each API key has its own private memory space
• Tags & filtering — organize memories by agent, session, or custom tags
• Zero config — just add your API key and go

License

MIT