Store and retrieve agent knowledge with persistent memory

agentMemory is a hybrid memory system built as an MCP server that gives AI agents persistent, searchable knowledge management. It extends an agent's capabilities with a memory bank that automatically syncs with a project's documentation, so context built up in one session stays available in later ones.

Use it when an agent needs to retain decisions, patterns, and findings across a project's lifetime instead of re-deriving them every session - for example, logging an architecture decision once and having a later session search for it by query, type, or tags. It only helps when the task matches the upstream skill's scope and the local project context; the skill itself warns not to treat its examples as a substitute for environment-specific tests, security review, or user approval before applying generated commands, dependencies, or changes with destructive or costly effects.

Setup requires Node.js, then installing and compiling the project before starting the MCP server with a project ID and the absolute path to the workspace:

npm install
npm run compile
npm run start-server <project_id> <absolute_path_to_workspace>

Once running, the server exposes four MCP tools: memory_search takes a query plus optional type and tags and returns matching memories; memory_write takes a key, type, content, and optional tags to record new knowledge or a decision; memory_read retrieves a specific memory by its key; memory_stats returns analytics on memory usage. On first run in a project it can also import existing markdown memory banks from .kilocode/, .clinerules/, or .roo/.

Writes made through the memory tools are automatically synced back to standard markdown files in the project, so the memory bank stays readable outside the MCP server too. Setup checks for an existing .agentMemory directory before installing.

Teams running AI coding agents that need a working loop of searching memory before a task, reading and writing during it, and recording new findings after it completes - so knowledge persists across sessions instead of resetting each time.