nowledge-mem

Nowledge Mem provides an MCP integration that lets your local Nowledge Mem knowledge graph interact with MCP clients through a standard streaming HTTP endpoint. This enables AI tools and agents to discover, capture, and search memories and insights from conversations and workflows. The server exposes an MCP endpoint at the URL configured in your environment (by default http://localhost:14242/mcp) and accepts connections from MCP clients via a streamableHttp interface. Use the MCP client headers to identify your application, and the system will route memory-related requests into Nowledge Mem’s knowledge graph for semantic search, graph exploration, and targeted memory retrieval. Tools enabled by this MCP server include semantic search across memories, graph-based exploration of entities and relationships, and autonomous agent access to curate and store memories during tasks. The integration supports capturing conversations, thread imports, and other interactions to enrich your personal knowledge graph while maintaining privacy-first guarantees.

Prerequisites:

• A supported runtime environment (Python with uvx or a container/orchestrator that can run your chosen command).
• Access to the MCP client app name you will pass via the APP header.

1. Install dependencies for the MCP runtime (choose one path):

• If using uvx (Python/uv): ensure uvx is installed and available on PATH.
• If using a container or system that supports your chosen command, ensure it can run the specified command and args.

2. Configure environment variables:

• Set MCP_URL to the MCP endpoint your clients will connect to (default http://localhost:14242/mcp).
• Set APP to a meaningful MCP Client App Name to identify your client in logs and dashboards.

3. Run the MCP server:

• For uvx based setup (as in this configuration): uvx nowledge-mem-mcp
• If you are deploying in a container/VM, ensure the image or binary includes the nowledge-mem-mcp entrypoint and that port 14242 (or your configured port) is exposed.

4. Verify the server is listening:

• curl -s http://localhost:14242/mcp/health
• You should receive a healthy status or a success payload from the MCP endpoint.

5. Connect an MCP client and test:

• Use the APP header to pass your client app name, e.g., APP: "MyMCPClient".
• Send a test memory query or store request to confirm proper integration.

Tips and common considerations:

• Ensure the MCP endpoint is reachable from the clients; open firewall rules and configure any reverse proxies accordingly.
• The APP header helps distinguish traffic from different clients in logs and analytics.
• For privacy and security, limit access to localhost in development and implement appropriate authentication/authorization in production.
• If you modify the mcp.json configuration on the client side, ensure the server recognizes changes and reloads the MCP connections as needed.
• If you encounter connectivity issues, verify that the MCP_URL matches the running server address and that the APP header is supplied by the client.
• This integration is designed to support memory capture, semantic search, and graph navigation; tune performance by adjusting graph algorithms and indexing strategies in Nowledge Mem as needed.