Claude Memory MCP
adeeljames/claude-memory-mcp
Memória persistente local para Claude — 70% menos tokens
Ver no GitHubServidor MCP que fornece memória persistente ao Claude entre sessões, reduzindo uso de tokens em 70% ao injetar apenas contexto relevante. Armazena e resume histórico de conversas localmente, com setup de um comando e zero latência via stdio.
Instalação
npx convez-ai adeeljames/claude-memory-mcp1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Manter contexto de projetos longos sem repetir histórico completo
- Reduzir custos de tokens em conversas recorrentes
- Armazenar conhecimento de domínio entre múltiplas sessões
Veredito

Solução prática e leve para quem usa Claude Desktop regularmente em projetos que se estendem por múltiplas conversas. A auto-summarização mantém o contexto eficiente, mas requer integração manual na config do Claude Desktop.
README
A tradução para o português deste README ainda está na fila.
🧠 Claude Memory MCP
A lightweight Model Context Protocol (MCP) server that gives Claude Desktop persistent memory across conversations. It stores, summarizes, and retrieves conversation history so Claude always remembers your context.
✨ Features
| Feature | Description |
|---|---|
| Persistent Memory | Saves every conversation turn to a local memory.json file |
| Auto-Summarization | Automatically compresses history after 10 turns to keep context lean |
| Fast Context Loading | Returns summary + last 3 turns on demand — no bloat |
| One-command Setup | Powered by uv — no virtualenv juggling needed |
| Zero Latency | Runs locally over stdio — no network calls |
🛠️ Tools Exposed
| Tool | Description |
|---|---|
get_context | Load compressed memory (summary + last 3 turns). Call at the start of every conversation. |
save_turn | Save one conversation turn. Call after every AI response. |
clear_memory | Wipe all stored memory and start fresh. |
🚀 Quick Start
Prerequisites
- Python 3.13+
uvinstalled
1. Clone & Install
git clone https://github.com/adeeljames/claude-memory-mcp.git
cd claude-memory-mcp
uv sync
2. Run the MCP Server (for testing)
uv run python server.py
3. Add to Claude Desktop
Open your Claude Desktop config file:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json
Add the following under mcpServers:
{
"mcpServers": {
"memory-mcp": {
"command": "uv",
"args": [
"run",
"--directory",
"/absolute/path/to/claude-mcp-optimize",
"python",
"server.py"
]
}
}
}
Replace
/absolute/path/to/claude-mcp-optimizewith the actual path on your machine.
Restart Claude Desktop — the memory-mcp server will appear in your tools list.
📁 Project Structure
claude-memory-mcp/
├── server.py # MCP server — all tools defined here
├── memory.json # Runtime memory file (auto-created, gitignored)
├── pyproject.toml # uv project config & dependencies
├── uv.lock # Locked dependency graph
└── README.md # You are here
⚙️ How It Works
Claude Desktop ──stdio──► server.py ──► memory.json
│
┌────────────────┘
│
get_context() → returns summary + last 3 turns
save_turn() → appends to history, triggers summary at 10 turns
clear_memory() → resets everything
🔧 Dependencies
| Package | Purpose |
|---|---|
mcp>=1.26.0 | Model Context Protocol SDK |
All dependencies are managed by uv and pinned in uv.lock.
📄 License
MIT — free to use, modify, and share.
Made with love by @muhammadadeelai