Memory Context MCP Server

A Model Context Protocol (MCP) server that gives LLMs persistent, semantic memory using vector search.

Built with Python, ChromaDB, and the official MCP SDK, this server allows AI models to store, search, update, and delete knowledge — and recall it across sessions using cosine-similarity vector search, not just keyword matching.

Features

• Vector Search — Semantic similarity search powered by sentence-transformers (runs 100% locally).

• MCP Protocol — Exposes store_memory, search_memory, update_memory, delete_memory as MCP tools.

• MCP Prompts — Slash-command style prompts (recall_topic, memory_summary, recall_session) for compatible LLM clients.

• Interactive CLI — A polished terminal interface with /commands for manual database management.

• Namespace Isolation — Memories are grouped by namespace to prevent cross-project leakage.

• Security — Size limits, tag validation, and local-only storage.

Related MCP server: Chroma MCP Server

Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Launch the CLI

python main.py

3. Launch as MCP Server (for LLM Clients)

python main.py --serve

CLI Commands

MCP Client Configuration (Claude Desktop)

Add this to your Claude Desktop claude_desktop_config.json:

{
  "mcpServers": {
    "memory-context-server": {
      "command": "python",
      "args": ["main.py", "--serve"],
      "cwd": "/path/to/MCP - SERVER"
    }
  }
}

Architecture

MCP - SERVER/
├── main.py                 # Entry point (CLI or --serve)
├── requirements.txt        # Python dependencies
├── .gitignore
├── README.md
├── chroma_data/            # ChromaDB persistent storage (auto-created)
└── src/
    ├── __init__.py
    ├── config.py           # Constants and security limits
    ├── db.py               # ChromaDB vector database layer
    ├── mcp_server.py       # MCP tools & prompts
    └── cli.py              # Interactive terminal CLI

License

MIT