Our tools are free. Help keep the research running.Buy us a coffee

Flagship · early access

MinniMemoryMCP: An MCP server that shrinks what Claude Code loads every turn

Your agent re-reads all its memory files on every turn, and most of it has nothing to do with the task at hand. MinniMemoryMCP checks those files, keeps what the agent needs every turn in a small always-on part (AlwaysOnMemory), and moves the rest into files it opens only when a task needs them (OnDemandMemory). Every write goes through your own approval, and nothing is lost.

$ npx -y minnimemory install

View on npm →

$ minnimemory init
  16 files load every turn
    CLAUDE.md, MEMORY.md + 14 linked files
  Open items, current state
    change often, not needed each turn
  Session logs
    history, only needed when asked
  Plan: 5,812 tokens load on demand
  Per turn: 12,511 -> 6,699 tokens (-47%)
$ minnimemory apply
  backed up every file it changes
  trimmed   14 files into AlwaysOnMemory
  moved     the rest to OnDemandMemory
  kept      every line, nothing lost
$ minnimemory undo
  restored  66 files it moved or changed
  removed   the files apply added

Simplified from real commands run on a full-size test memory. The numbers are the tool’s own and count what loads each turn.

Demo

Less for Claude Code to re-read every turn

Only the small AlwaysOnMemory part loads every turn. The rest waits in OnDemandMemory files the agent opens when a task needs them.

Nothing in your memory is lost

Every line is kept, the originals are backed up first, and undo puts them back. In Claude Code the server only returns the command to run; your agent runs it through your own approval, so you see every change.

Plain Markdown files you own

It works on plain text files you already own, and the result is plain Markdown, so your memory keeps working when you switch models. It connects to Claude Code through MCP, the standard way AI tools connect to add-ons.

Tools: Four commands. init runs first: it maps what your agent loads and previews the split. apply makes the split, or redoes it after you edit. undo restores the originals. doctor re-checks your memory later. In Claude Code, apply and undo return a command your agent runs through your approval.
Install: Run npx -y minnimemory install to add it to Claude Code, then ask your agent to run minnimemory init on your memory.
Works with: Claude Code. Other tools are not supported yet; the files it writes are plain Markdown any agent can read.
Runs: Entirely on your computer. There is no AI model inside it, no API key, and the tools send nothing over the internet. Every line is kept.
Applies: MinniMemory Interface v1.0, the same rules our own agents run on.
Status: Early access, on npm as minnimemory. New releases ship as rules are added.
Pricing: Free during early access. Launch pricing is not set yet; the rule set itself stays proprietary.

What we measured

On a full-size test memory (CLAUDE.md, MEMORY.md and 14 linked files), the part loaded every turn drops from 12,511 to 6,699 tokens. That counts what loads each turn.

What it is not

Not a database, not a search add-on, and not a hosted memory you lose when you switch tools. It works on plain text files you already own, backs up the originals first, and leaves plain Markdown behind, so your memory keeps working when the model changes.

Talk to us before it gets expensive.

Send us what your agent loads each turn. We apply the interface and report tokens saved against answer quality.