🧠 MCP Server for MnemoX Lite v1.1.1 - Semantic Memory for LLMs via Model Context Protocol
Connect Claude Desktop (and other MCP-compatible LLMs) to your MnemoX Lite semantic memory system with native Google Cloud vector search.## 🚀 Quick Start### 1. Run with NPX (Recommended)bash# Interactive setup with auto-detectionnpx @mnemox/mcp-server-lite# Direct server mode npx @mnemox/mcp-server-lite --server
The setup wizard will generate a configuration like this for your claude_desktop_config.json
:
{
"mcpServers": {
"mnemox-lite": {
"command": "npx",
"args": ["mnemox-mcp-server", "--server"],
"env": {
"MNEMOX_API_URL": "http://localhost:3000/api/mcp",
"MNEMOX_API_KEY": "your-api-key",
"MNEMOX_USER_ID": "your-user-id"
}
}
}
}
After adding the configuration, restart Claude Desktop. The MCP server will start automatically when you use Claude.
Once connected, Claude will have access to these memory tools:
Store information in semantic memory for future reference.
store_information({
content: "React hooks allow functional components to use state",
tags: ["react", "hooks", "frontend"],
source: "learning_session"
})
Search semantic memory for relevant information.
search_memory({
query: "React hooks state management",
maxResults: 10,
similarityThreshold: 0.7
})
Find information related to a specific memory fragment.
relate_information({
fragmentId: "fragment-123",
maxResults: 10
})
Organize information into thematic contexts.
contextualize_information({
query: "React development best practices",
createIfNotExists: true,
maxFragments: 50
})
Get statistics about your semantic memory usage.
get_memory_stats()
-
MNEMOX_API_URL
: URL of your MnemoX Lite server (default:http://localhost:3000/api/mcp
) -
MNEMOX_API_KEY
: API key for authentication (default:demo-key
) -
MNEMOX_USER_ID
: User ID for memory operations (default:demo-user
)
-
macOS:
~/Library/Application Support/Claude/claude_desktop_config.json
-
Windows:
%APPDATA%\Claude\claude_desktop_config.json
Test the server manually:
# Start the MCP server
npx mnemox-mcp-server --server
# In another terminal, test with MCP client tools
# (Requires MCP development tools)
# Clone and install
git clone <repository>
cd mcp-server
npm install
# Run locally
npm start
mcp-server/
├── bin/
│ └── mnemox-mcp # NPX executable
├── src/
│ └── server.js # MCP server implementation
├── package.json
└── README.md
Human: Remember that I prefer TypeScript over JavaScript for new projects.
Claude: I'll store that preference for you.
[Uses store_information tool]
✅ Information stored successfully!
Fragment ID: frag_abc123
Content: User prefers TypeScript over JavaScript for new projects
Human: What programming language should I use for my new web project?
Claude: Let me check your preferences first.
[Uses search_memory tool]
🔍 Found 1 relevant memory:
1. [Score: 0.85] User prefers TypeScript over JavaScript for new projects
Based on your stored preference, I recommend using TypeScript for your new web project...
Human: I've been learning about React hooks. Can you organize all my React knowledge?
Claude: I'll organize your React knowledge into a context.
[Uses contextualize_information tool]
📚 Context "React development knowledge" created with 15 fragments
Your React knowledge has been organized! This includes:
- React hooks concepts
- Component patterns
- State management
- Performance optimization tips
❌ "API Error: 401 Unauthorized"
- Check your
MNEMOX_API_KEY
in the configuration - Ensure your MnemoX server is running and accepting the API key
❌ "Error calling MnemoX API: fetch failed"
- Verify your
MNEMOX_API_URL
is correct - Check that your MnemoX server is running on the specified URL
- Try accessing the health endpoint:
curl http://localhost:3000/api/mcp/health
❌ "NPX package not found"
- Ensure you have Node.js 18+ installed
- Try clearing NPX cache:
npx clear-npx-cache
Enable verbose logging:
# Set debug environment variable
MNEMOX_DEBUG=1 npx mnemox-mcp-server --server
MIT License - see LICENSE file for details.
🧠 Built with ❤️ by the MnemoX Team