Simple Memory Extension

🚀 Simple Memory Extension MCP Server

A Memory Context Proxy (MCP) server designed to expand the agent's context window and memory capabilities. It's incredibly useful during large - scale feature development or vibe coding, allowing you to store, recall progress, critical moments, changes, or any other noteworthy information. Agents can store memories, retrieve them as needed, or manage their memories autonomously (through cursor rules, etc.).

🚀 Quick Start

Starting the Server

npm install
npm start

Available Tools

Context Item Management

• store_context_item - Store a value and key in a namespace.
• retrieve_context_item_by_key - Retrieve a value by its key.
• delete_context_item - Delete a key - value pair.

Namespace Management

• create_namespace - Create a new namespace.
• delete_namespace - Delete a namespace and all its contents.
• list_namespaces - List all namespaces.
• list_context_item_keys - List keys in a namespace.

Semantic Search

• retrieve_context_items_by_semantic_search - Find items based on their meaning.

Semantic Search Implementation

1. Transform the query into a vector using the E5 model.
2. Automatically split the text into paragraphs to enhance matching.
3. Calculate the cosine similarity between the query and stored paragraphs.
4. Filter the results based on a threshold and sort them by similarity.
5. Return the top - matching items and their full values.

💻 Usage Examples

Basic Usage

# Starting the server
npm install
npm start

Advanced Usage

# Development server
npm run dev

# Formatting code
npm run format

📚 Documentation

.env Configuration

# SQLite database file path
DB_PATH=./data/context.db

PORT=3000

# Use HTTP SSE or Stdio
USE_HTTP_SSE=true

# Log configuration: debug, info, warn, error
LOG_LEVEL=info

Semantic Search

This project includes a semantic search feature using the E5 embedding model from Hugging Face. This enables you to find context items based on their meaning rather than just by keys.

Setup

The semantic search feature requires Python dependencies, which should be automatically installed when you run npm run start.

Embedding Model

We are using [intfloat/multilingual - e5 - large - instruct](https://huggingface.co/intfloat/multilingual - e5 - large - instruct)

Note

This project was mainly developed during vibe coding, so don't expect too much :D. But it does work, and I find it useful. Feel free to contribute or offer suggestions.