agent-memory-mcp

Run the agent-memory MCP server against your workspace, then use memory_write to store decisions and patterns with keys and tags. Later sessions retrieve them with memory_search or memory_read, so knowledge persists beyond a single conversation.

Use the memory_search tool with a query string plus optional type and tags filters. For example, searching with type set to pattern returns only pattern entries matching your query, such as authentication patterns.

You need Node.js version 18 or later. Clone the agentMemory repository, run npm install and npm run compile, then start the server with npm run start-server passing a project ID and the absolute path to your target workspace.

Yes. The memory_stats tool returns analytics on memory usage, and a standalone dashboard can be launched with npm run start-dashboard pointing at your workspace. The dashboard is accessible at http://localhost:3333.

The memory bank stores typed entries such as architecture notes, design patterns, and decisions, each with a key, content, and optional tags. Entries are written with memory_write and retrieved individually by key using memory_read.