mcp -memories-off

This MCP server provides a lightweight, local memory knowledge base. It processes documents, conversations, and notes to extract entities, relations, and observations, building a personal knowledge graph that can be queried and reused by your LLM. You can manage entities and relationships, search the graph, and backup or inspect the knowledge base. The server exposes a set of tools for creating/updating entities and relations, retrieving graph information, performing searches, and performing maintenance tasks such as backups. Typical usage is to run the server locally and interact with it through MCP clients using the memory-focused commands and endpoints.

You can manage knowledge with the provided toolset: upsert_entities to create or update multiple entities; create_relations to define relationships; upsert_observations to add observations to entities; delete_entities and delete_observations for cleanup; read_graph and read_subgraph to inspect the knowledge graph; search_nodes_anywhere and search_nodes_smart for retrieval; and backup_graph for safeguarding your data. It also supports organizational features like list_entity_types, list_relation_types, and rename_entity to adjust naming. All of these operations are designed to work offline on local files, ensuring privacy and portability.

Prerequisites:

• Node.js (LTS) installed on your machine
• npm (comes with Node.js) or corepack/yarn as preferred

Installation steps:

1. Prepare the environment variables (adjust paths as needed):

   • MEM_NAME: memory
   • MEM_PATH: Path to the knowledge graph file, e.g. $HOME/mcp-server-memories-off.yaml
   • MEM_LOG_DIR: Directory for logs, e.g. /tmp

2. Install and run the server using npx (no global install required):

   • npx -y mcp-server-memories-off

3. Alternatively, run via a global install (if you prefer):

   • npm install -g mcp-server-memories-off
   • mcp-server-memories-off

4. If you plan to customize the configuration programmatically, create a mcp.json (or equivalent) and point to the same memory configuration: { "mcpServers": { "memories-off": { "command": "npx", "args": ["-y", "mcp-server-memories-off"], "env": { "MEM_PATH": "$HOME/mcp-server-memories-off.yaml" } } } }

Prerequisites recap:

• Node.js installed
• Access to npm (or yarn/bun depending on workflow)
• Optional: Bun for build steps mentioned in the repo (bun run build) if you clone and build locally.

Tips and common notes:

• The server stores data locally in a YAML/graph file; ensure MEM_PATH points to a writable location with sufficient disk space.
• MEM_LOG_DIR can be set to a dedicated logs folder to avoid cluttering system temp directories.
• The default command to start is npx -y mcp-server-memories-off, which ensures the latest published package is used without a global install.
• When upgrading the package, re-run the start command to load the latest code and schema.
• If you encounter token limits during large graph reads, prefer read_subgraph or read_graph_manual for structured retrieval.
• For local development, you can build and run via bun: bun run build followed by node dist/index.js, per the project notes.
• Environment isolation (e.g., using a dedicated project directory) helps prevent path confusion across multiple MCP servers.