CLI Tool

Put your MEMORY.md on a diet.

Analyzes your memory files, generates a machine-readable dispatch table, and shows where your context budget goes. 98% token savings on a real 31-topic workspace.

Get started Read the Handbook

Install

npm install -g @mcptoolshop/claude-memories

Analyze

claude-memories analyze MEMORY.md

Index

claude-memories index MEMORY.md --lazy

Features

Why your agent needs a memory index.

Context savings

Turns 40K+ tokens of always-loaded memories into an on-demand dispatch table. Only the relevant topic loads per task.

Auto-extraction

Keywords are extracted from topic names and headings automatically. Optional frontmatter gives fine-grained control.

Full diagnostics

Token budget dashboard, orphan detection, duplicate references, structural validation — all from one CLI.

Usage

Install globally

npm install -g @mcptoolshop/claude-memories

Analyze structure

claude-memories analyze MEMORY.md
# Shows topics, references, token costs

Generate dispatch table

claude-memories index MEMORY.md --lazy
# Writes index.json for ai-loadout

Token budget dashboard

claude-memories stats MEMORY.md
# Total tokens, savings %, per-topic breakdown

Architecture

Layer 2 adapter in the Knowledge OS stack.

Kernel

@mcptoolshop/ai-loadout provides routing types, matching, and validation — the shared foundation.

Adapter

claude-memories converts MEMORY.md into a LoadoutIndex. Same kernel, different document type.

Zero dependencies

Local-only, no network calls, no telemetry. Deterministic: same inputs always produce the same outputs.