Simple Memory Claude

simple-memory-claude

Sistema de memoria personal simple y efectivo para Claude Code. Simple and effective personal memory system for Claude Code.

[](https://nodejs.org) [](LICENSE) [](https://modelcontextprotocol.io)

---

¿Qué es esto? / What is this?

[ES] Un servidor MCP que le da memoria persistente a Claude Code. Tus notas viven en archivos Markdown simples, sincronizados en la nube. Claude puede leerlas, crearlas, buscarlas y organizarlas durante cualquier sesión de trabajo.

[EN] An MCP server that gives Claude Code persistent memory. Your notes live in plain Markdown files, synced to the cloud. Claude can read, create, search and organize them during any working session.

---

¿Por qué este sistema? / Why this system?

[ES]

Hay muchos sistemas de memoria para Claude y agentes de IA, pero la mayoría son complejos, requieren configuraciones largas o dependen de servicios externos. Este sistema parte de una premisa diferente:

• Simple: un solo archivo de servidor (index.js), una dependencia
• Tuyo: las notas son archivos .md en tu disco — sin bases de datos, sin APIs externas
• Portátil: funciona con iCloud, OneDrive, Google Drive, Dropbox o cualquier carpeta local
• Transparente: puedes abrir y editar tus notas en cualquier editor de texto

[EN]

There are many memory systems for Claude and AI agents, but most are complex, require lengthy setup or depend on external services. This system starts from a different premise:

• Simple: a single server file (index.js), one dependency
• Yours: notes are .md files on your disk — no databases, no external APIs
• Portable: works with iCloud, OneDrive, Google Drive, Dropbox or any local folder
• Transparent: you can open and edit your notes in any text editor

---

Características / Features

• 29 MCP tools in four blocks: base CRUD, targeted editing, wikilink/tag maintenance, cheap reads
• Targeted editing — find/replace, append, prepend, section update without re-sending the whole file
• Markdown-aware section operations — read or rewrite a section by its heading
• Cheap reads — peek (frontmatter + first paragraph), read a single section, read frontmatter only
• Wikilink tools — list broken links, find backlinks, find orphans, rename a wikilink globally
• Tag tools — list all #snake_case hashtags with note counts
• Vault maintenance — recently updated, validate note structure, bulk move, move category
• Optional authorship tracking — Markdown Annotations support (opt-in via KB_ENABLE_ANNOTATIONS=1). Tracks which author wrote which ranges of each note, preserves human authorship when Claude edits afterward. Renders natively in iA Writer.
• AND logic search — all terms must appear in the note
• Automatic wikilink update when a note is renamed
• Frontmatter with title, category, tags, created and updated dates
• Broken wikilink detection on save
• Configurable vault path via environment variable

---

Estructura del proyecto / Project structure

simple-memory-claude/
├── server/
│   ├── index.js          ← MCP server (the only file that runs)
│   └── package.json
├── vault-example/        ← ready-to-use example vault
│   ├── HOME.md           ← start here — customize with your info
│   ├── clientes/         ← client profiles and context
│   ├── stack/            ← tools and services you use
│   ├── referencias/      ← technical guides and references
│   ├── proyectos/        ← your projects
│   └── problemas-resueltos/ ← solved problems worth remembering
└── docs/
    ├── mac-icloud.md     ← installation guide for Mac + iCloud
    ├── windows.md        ← installation guide for Windows
    ├── linux.md          ← installation guide for Linux
    └── tools-reference.md ← complete reference for all 9 tools

---

Instalación rápida / Quick start

Requisitos / Requirements

• Node.js 18 or later
• Claude Code (npm install -g @anthropic-ai/claude-code)
• A cloud sync folder (iCloud, OneDrive, Google Drive, Dropbox) — or any local folder

1. Clonar el repositorio / Clone the repository

git clone https://github.com/jmpdsevilla/simple-memory-claude.git
cd simple-memory-claude/server
npm install

2. Crear tu bóveda / Create your vault

Copy the example vault to your cloud folder and customize HOME.md:

Mac + iCloud:

cp -r vault-example ~/Library/Mobile\ Documents/com~apple~CloudDocs/my-memory

Windows + OneDrive:

xcopy /E /I vault-example "%USERPROFILE%\OneDrive\my-memory"

Linux + Dropbox:

cp -r vault-example ~/Dropbox/my-memory

3. Configurar Claude Code / Configure Claude Code

Mac + iCloud (no env var needed — works out of the box):

{
  "mcpServers": {
    "simple-memory": {
      "command": "node",
      "args": ["/absolute/path/to/simple-memory-claude/server/index.js"]
    }
  }
}

Windows / Linux / custom path:

{
  "mcpServers": {
    "simple-memory": {
      "command": "node",
      "args": ["/absolute/path/to/simple-memory-claude/server/index.js"],
      "env": {
        "MEMORY_PATH": "/absolute/path/to/your/vault"
      }
    }
  }
}

4. Verificar / Verify

Restart Claude Code and ask Claude to run get_index. You should see your vault index.

---

Authorship tracking (optional)

[ES] El MCP soporta opcionalmente Markdown Annotations, una spec abierta originalmente de iA Writer que permite registrar qué autor escribió qué parte de cada nota. Cuando se activa, las notas escritas por Claude llevan al final un bloque que atribuye el cuerpo a &Claude; cuando un humano edita la nota en iA Writer, sus rangos quedan marcados como @nombre, y la siguiente vez que el MCP toque la nota preserva esa autoría sin sobrescribirla.

[EN] The MCP optionally supports Markdown Annotations, an open spec originally from iA Writer that records which author wrote which part of each note. When enabled, notes written by Claude carry a block at the end attributing the body to &Claude; when a human edits the note in iA Writer, their ranges are marked as @name, and the next time the MCP touches the note their authorship is preserved instead of overwritten.

How to enable / Cómo activarlo:

Add KB_ENABLE_ANNOTATIONS=1 to the MCP environment:

{
  "mcpServers": {
    "simple-memory": {
      "command": "node",
      "args": ["/absolute/path/to/simple-memory-claude/server/index.js"],
      "env": {
        "KB_ENABLE_ANNOTATIONS": "1"
      }
    }
  }
}

Two extra tools become available / Dos herramientas adicionales:

• read_authorship(name) — compact summary of who wrote which ranges.
• migrate_annotations({ dry_run }) — one-shot tool to add the block to all existing notes (preserves the original updated field). Run with dry_run: true first to audit.

Author name / Nombre del autor:

By default annotations are signed as Claude (noreply@anthropic.com). Override the signature with KB_AUTHOR_NAME and KB_AUTHOR_EMAIL so each server instance signs with its own identity (useful when several assistants share one vault). Por defecto las anotaciones se firman como Claude (noreply@anthropic.com). Cambia la firma con KB_AUTHOR_NAME y KB_AUTHOR_EMAIL para que cada instancia del servidor firme con su propia identidad (útil cuando varios asistentes comparten una misma bóveda).

Important / Importante:

• Disabled by default. Without KB_ENABLE_ANNOTATIONS=1 the MCP behaves identically to v1.2.0: no annotation block at the end of files.
• Editors that don't support the spec render the block as plain text at the end of the file. Only enable if you use iA Writer or another compatible editor.

---

Estructura recomendada de la bóveda / Recommended vault structure

my-memory/
├── HOME.md                         ← read at the start of every session
├── clients/
│   └── client-name/
│       └── profile.md
├── stack/                          ← tools, APIs, credentials
├── projects/                       ← your own projects
├── references/                     ← guides, patterns, resources
├── infrastructure/                 ← servers, DNS, deployments
└── solved-problems/
    └── area/
        └── solution.md

The folder structure is flexible — create the categories that make sense for your work.

---

Las herramientas / The tools

(29 tools in total — 27 always available + 2 unlocked by KB_ENABLE_ANNOTATIONS=1)

Base CRUD (9)

| Tool | What it does | |---|---| | write_note | Create or update a note (full upsert) | | read_note | Read a note (searches all categories) | | search_notes | Search by free text (AND logic) | | list_notes | List notes, filtered by category or tag | | get_index | Full vault index with backlinks | | delete_note | Delete a note (warns about backlinks) | | create_category | Create a new folder | | move_note | Move/rename a note (auto-updates wikilinks) | | delete_category | Delete an empty folder |

Targeted editing (5)

| Tool | What it does | |---|---| | edit_note | find/replace inside a note (fails on ambiguity unless replace_all) | | append_to_note | Append content at the end (before trailing hashtags) | | prepend_to_note | Insert content at the beginning (after h1) | | update_section | Replace a markdown section by its heading | | insert_after_section | Insert a new section right after another |

Wikilink and tag maintenance (6)

| Tool | What it does | |---|---| | list_broken_links | All broken wikilinks in the vault | | find_backlinks | Backlinks of a note (without loading content) | | find_orphans | Notes with no backlinks and no outlinks | | rename_wikilink | Globally substitute [[old]] with [[new]] | | list_tags | All #snake_case hashtags with usage counts | | update_frontmatter | Update YAML fields without touching the body |

Cheap reads (3)

| Tool | What it does | |---|---| | peek_note | Frontmatter + first paragraph | | read_section | Just one markdown section | | read_frontmatter | Just the YAML |

Vault maintenance (4)

| Tool | What it does | |---|---| | recently_updated | Notes modified in the last N days | | move_category | Rename a folder, updates each note frontmatter | | validate_note | Check frontmatter, hashtags, See also, broken links | | bulk_move | Move several notes to the same category |

See [docs/tools-reference.md](docs/tools-reference.md) for the full reference.

---

Protocolo de uso / Usage protocol

Add this instruction to your Claude Code settings or CLAUDE.md to get the most out of the system:

At the start of each session: read_note("HOME"), then get_index if needed.
Save automatically: credentials, solutions, configurations, commands worth remembering.
Use [[slug]] wikilinks between notes.
Every note should end with a "See also" section with 2–5 wikilinks.

---

Wikilinks

Notes link to each other using the [[slug]] format:

## See also

- [[supabase]] — database used in this project
- [[acme-corp]] — client this belongs to

Rules:

• Use the filename slug (kebab-case, no .md)
• No paths: ~~[[stack/supabase]]~~ → [[supabase]]
• No aliases: ~~[[supabase|Database Config]]~~ → [[supabase]]

---

Guías de instalación detalladas / Detailed installation guides

• [Mac + iCloud Drive](docs/mac-icloud.md)
• [Windows + OneDrive / Google Drive](docs/windows.md)
• [Linux + Dropbox / Google Drive](docs/linux.md)

---

Autor / Author

Creado por José Manuel Pérez, fundador de santa marta crea — Agencia Creativa Digital. Santa Marta, Colombia.

Created by José Manuel Pérez, founder of santa marta crea — Creative Digital Agency. Santa Marta, Colombia.

• Web: santamartacrea.com
• GitHub: @jmpdsevilla

---

Historial de cambios / Changelog

Ver [CHANGELOG.md](CHANGELOG.md) para el historial completo de versiones. See [CHANGELOG.md](CHANGELOG.md) for the full version history.

---

Licencia / License

[MIT](LICENSE) — free to use, modify and distribute.

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jmpdsevilla
• Source: jmpdsevilla/simple-memory-claude
• License: MIT

Install and usage instructions live in the source repository linked above.