mnemox-mcp

Mnemox MCP server — score prompts, count tokens, save and retrieve memories.

Works with Claude Code, Cowork, Cursor, and any MCP-compatible AI tool.

Install

npm install -g mnemox-mcp

Or run directly without installing:

npx mnemox-mcp

Tools

| Tool | Description | |------|-------------| | score_prompt | Score a prompt 0-100 using 8 rules. 100% local, no API. | | count_tokens | Estimate token count + cost for any model. 100% local. | | get_memory | Retrieve or semantically search saved memories. | | save_memory | Save a memory to the Mnemox backend. |

Configure in Claude Code

Add to ~/.claude/claude_desktop_config.json:

{
  "mcpServers": {
    "mnemox": {
      "command": "mnemox-mcp",
      "env": {
        "MNEMOX_API_URL": "https://mnemox-production.up.railway.app"
      }
    }
  }
}

Or with npx (no global install needed):

{
  "mcpServers": {
    "mnemox": {
      "command": "npx",
      "args": ["mnemox-mcp"],
      "env": {
        "MNEMOX_API_URL": "https://mnemox-production.up.railway.app"
      }
    }
  }
}

Configure in Cursor

Add to Cursor settings → MCP:

{
  "mnemox": {
    "command": "mnemox-mcp"
  }
}

Environment Variables

| Variable | Default | Description | |----------|---------|-------------| | MNEMOX_API_URL | https://mnemox-production.up.railway.app | Mnemox backend URL |

score_prompt and count_tokens are 100% local — they work without any backend connection.

get_memory and save_memory require the backend to be reachable.

Example Usage

Once configured, ask your AI assistant:

• "Score this prompt: [your prompt]"
• "How many tokens is this?"
• "Save a memory: I prefer concise bullet-point answers"
• "What do I have saved in Mnemox?"