# MCP Memory Server

## Overview
This repository contains a simple **MCP (Model Context Protocol)** server that provides basic key‑value memory operations with optional namespace support. The server is built on top of the `fastmcp` framework and stores data in a JSON file.

The project also includes a small client script demonstrating how to interact with the server via MCP stdio transport.

## File Structure
- **memory_server.py** – Implements the MCP server with tools: `save`, `get`, `delete`, `list_keys`, `save_with_namespace`, `get_by_namespace`.
- **memory_client.py** – Example client that connects to the server and performs a few operations.
- **requirements.txt** – Python dependencies.
- **README.md** – Project documentation.

## Installation
```bash
pip install -r requirements.txt
```

## Running the Server
```bash
python memory_server.py
```
The server listens on stdio transport and can be used by any MCP client.

## Using the Client
```bash
python memory_client.py
```
This script demonstrates:
1. Saving a value with namespace `default`.
2. Retrieving all items in that namespace.
3. Listing keys matching a pattern.

The expected output is similar to:
```
Сохранено: True
Данные namespace 'default':
  username: Алексей
Ключи с 'name': ['default:user_name']
```

## MCP Tool Details
| Tool | Purpose |
|------|---------|
| `save` | Store a value under a key. |
| `get` | Retrieve the stored item for a key. |
| `delete` | Remove a key from storage. |
| `list_keys` | List all keys with optional glob pattern. |
| `save_with_namespace` | Store a value under ``namespace:key``. |
| `get_by_namespace` | Return all items belonging to a namespace. |

## Notes
- Data is persisted in `memory_data.json` located next to the script.
- Timestamps are stored in UTC ISO format.
- The server uses atomic writes for safety.
- Wildcard patterns use Unix shell syntax via `fnmatch`.

---

Feel free to extend or adapt this example for your own multi‑agent systems.