mem-brain-mcp 1.0.0

MCP Server for Mem-Brain API - Exposes memory operations as MCP tools

Mem-Brain MCP Server

MCP (Model Context Protocol) server that exposes Mem-Brain API functionality as standardized tools for AI agents. Built with FastMCP for production-ready HTTP/SSE transport.

Features

• Memory Management: Create, read, update, and delete memories
• Semantic Search: Search memories using vector similarity
• Graph Operations: Find paths and neighborhoods in the memory graph
• Statistics: Get insights about your memory system
• Link Management: Link and unlink memories
• HTTP/SSE Transport: Run independently, accessible remotely
• CLI Interface: Packaged for easy global execution

Instant Execution (Recommended)

You can run the MCP server instantly without manual installation using uv:

# Run using uvx (loads environment from .env or shell)
uvx mem-brain-mcp

# Run with custom API URL
export API_BASE_URL=http://your-api-alb-url.com
uvx mem-brain-mcp

Installation

From Source

1. Install using uv (recommended) or pip:

cd mem-brain-mcp
uv pip install .

2. Run globally:

mem-brain-mcp

Configuration

The server reads configuration from environment variables or a .env file in the current working directory:

# API Configuration
API_BASE_URL=http://localhost:8000
# NOTE: API_KEY is optional here - per-user API keys are configured in MCP clients
API_KEY=your_api_key_here  # Optional: fallback for single-user scenarios

# MCP Server Configuration
MCP_SERVER_HOST=0.0.0.0
MCP_SERVER_PORT=8100

# Logging
LOG_LEVEL=INFO

Per-User API Key Configuration

Each user must configure their own API key in their MCP client for proper user isolation. The server extracts tokens from request headers.

Cursor IDE (~/.cursor/mcp.json)

{
  "mcpServers": {
    "mem-brain": {
      "url": "http://localhost:8100/mcp",
      "headers": {
        "Authorization": "Bearer your-jwt-token"
      }
    }
  }
}

Claude Desktop

Option 1: Native Remote (Pro/Team plans)

Add to your Claude Desktop configuration (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "mem-brain": {
      "url": "http://your-deployed-url/mcp",
      "headers": {
        "Authorization": "Bearer your-jwt-token"
      }
    }
  }
}

Option 2: via CLI (Any plan)

{
  "mcpServers": {
    "mem-brain": {
      "command": "uvx",
      "args": ["mem-brain-mcp"],
      "env": {
        "API_BASE_URL": "http://your-deployed-url",
        "JWT_SECRET_KEY": "your-secret"
      }
    }
  }
}

AWS ECS Deployment

The MCP server can be deployed to AWS ECS (Fargate) with an Application Load Balancer.

Quick Start

1. Set up security groups (see aws/security-groups.md)
2. Deploy using the script:

cd mem-brain-mcp/aws
./deploy.sh ap-south-1 membrain-mcp membrain-cluster membrain-mcp

For detailed instructions, see aws/DEPLOYMENT.md.

Development

Running Tests

pytest

Build and Publish

# Build the wheel
python3 -m build

# Upload to PyPI
twine upload dist/*

License

Same as Mem-Brain API project.