agentMemory

agentMemory acts as a hybrid memory management MCP server that augments AI agents with a structured, searchable memory database. It sits between built-in memory banks like KiloCode, Cline, and RooCode and your agent workflows, offering fast queries, analytics, and a bi-directional sync that keeps both the Markdown memory banks and the JSON database in sync. Tools exposed by the MCP server include memory_read, memory_write, memory_search, memory_list, memory_update, and memory_stats, enabling agents to read specific memories, append new notes, search by content or tags, and view usage analytics. The system is designed to work across multiple projects, providing cross-project memory accessibility and consistent workflow guidance for memory-driven decisions.

Prerequisites:

• Node.js (LTS) installed on your machine
• Git installed
• Basic command-line familiarity

Installation steps:

1. Clone the repository or set up your project workspace:

   git clone https://github.com/yourusername/agentMemory cd agentMemory

2. Install dependencies and build the project:

   npm install npm run compile npm run package

3. Start the MCP server (example):

   node server.js

4. Connect an MCP client or configure your MCP runner to use the following configuration (see mcp_config in this doc):

   Ensure the server is reachable at the configured port (default 3000).

5. If you’re packaging as a VS Code extension, follow the marketplace installation flow described in the README to enable the extension and its MCP server integration.

Tips and common considerations:

• Ensure your Markdown memory banks (.kilocode, .clinerules, .roo) are readable by the agentMemory bi-directional sync process to enable automatic importing and updating.
• Use memory_search with explicit types and tags to improve precision, e.g., memory_search({ query: 'OAuth', type: 'pattern', tags: ['security'] }).
• Regularly check the dashboard to monitor memory trends and agent activity for changes or anomalies.
• If cross-project search is disabled or not yielding results, verify cross-project indices are built and that permissions allow cross-project queries.
• Environment variables such as PORT and any authentication tokens should be secured; avoid hard-coding sensitive values in public repos.
• When upgrading, re-run the MCP packaging steps to refresh the extension or server binary and ensure compatibility with the latest memory schema.