agentic-recall

A local MCP server that gives an assistant long-term memory for agentic tasks by reading markdown memory files, and says when it doesn't know.

What you can do

Provides long-term memory for agentic tasks, backed by a folder of markdown memory files on the user's machine, so an assistant can recall earlier context across sessions. Reports when it does not know rather than guessing. Keeps an embedding model cache, vector cache and indexes in a separate memory root. The manifest lists no individual tools, so the exact operations are not documented here.

When to use it

Worth adding when you want an assistant to carry persistent, file-based memory between agentic tasks and to admit gaps instead of inventing answers. Suited to users who already keep their notes or memories as markdown files they control.

Requirements

Runs as a local stdio process installed from the npm package agentic-recall, so Node.js is needed. Requires the MEMORY_DIR environment variable set to an absolute path of the folder holding your markdown memories; the server never searches the disk for it. Optional MEMORY_ROOT sets where caches and indexes are stored. Desktop only.