@sylphlab/tools-memory-mcp

MCP Server adaptor for the Knowledge Graph memory tools.

Overview

This package provides an MCP (Model Context Protocol) server that exposes the knowledge graph memory management tools defined in @sylphlab/tools-memory. It acts as an adaptor, making the core logic available to AI agents via the MCP protocol over stdio.

The server reads/writes the knowledge graph data (default: memory.jsonl) relative to the current working directory (process.cwd()) where the server is started.

Features

• MCP Server: Runs as a stdio-based MCP server.
• Adaptor: Connects core @sylphlab/tools-memory tools to the MCP protocol using @sylphlab/tools-adaptor-mcp.
• CWD-based Storage: Inherits the CWD-based storage behavior from @sylphlab/tools-memory.
• Exposed Tools: Provides MCP access to createEntities, createRelations, addObservations, deleteEntities, deleteObservations, deleteRelations, readGraph, searchNodes, openNodes.

Usage

1. Installation:

   # Usually installed as part of the main project or globally
   pnpm add @sylphlab/tools-memory-mcp

2. Running the Server: The package provides a binary that can be run directly:

   # Directly (ensure dependencies are built)
   node ./node_modules/.bin/mcp-memory
   
   # Or via pnpm from the workspace root (if linked correctly)
   # pnpm --filter @sylphlab/tools-memory-mcp start

   The server will start and listen on stdio. It will create/use memory.jsonl in the directory where you run the command.

3. Environment Variable (Optional): You can specify a different file path or name relative to the CWD using the MEMORY_FILE_PATH environment variable (this is handled by the underlying @sylphlab/tools-memory package):

   MEMORY_FILE_PATH=./data/my_memory.jsonl node ./node_modules/.bin/mcp-memory

Development

• Build: pnpm build (Builds this package and its dependencies)
• Watch: pnpm dev
• Lint/Format: pnpm format, pnpm check

License

MIT