MCP Memory Server

MCP Memory Server

A streamable HTTPS (SSE) MCP server with personal memory layer using AWS ECS Fargate, RDS PostgreSQL, and OpenSearch Serverless for vector search.

Features

• MCP Tools: find_person, get_profile, remember_note, semantic_search
• SSE Transport: Streaming responses over HTTPS
• Relational Storage: AWS RDS PostgreSQL for structured data
• Vector Search: OpenSearch Serverless with k-NN for semantic search
• Embeddings: AWS Bedrock Titan for text embeddings
• Secure: Bearer token auth, VPC isolation, IAM roles

Quick Start

Local Development

1. Install dependencies:

npm install

1. Start PostgreSQL:

docker-compose up -d postgres

1. Run migrations:

docker exec -i mcp-postgres psql -U postgres -d memory < sql/001_schema.sql
docker exec -i mcp-postgres psql -U postgres -d memory < sql/002_seed_data.sql

1. Configure environment:

cp .env.example .env
# Edit .env with your AWS credentials

1. Start dev server:

npm run dev

AWS Deployment

1. Configure AWS SSO:

aws sso login --profile leogao

1. Build and push Docker image:

# Get AWS account ID
AWS_ACCOUNT_ID=$(aws sts get-caller-identity --query Account --output text)
AWS_REGION=us-east-1

# Build image
docker build -t mcp-memory:latest .

# Login to ECR
aws ecr get-login-password --region $AWS_REGION | docker login --username AWS --password-stdin $AWS_ACCOUNT_ID.dkr.ecr.$AWS_REGION.amazonaws.com

# Create repository (if needed)
aws ecr create-repository --repository-name mcp-memory-prod-app --region $AWS_REGION

# Push image
docker tag mcp-memory:latest $AWS_ACCOUNT_ID.dkr.ecr.$AWS_REGION.amazonaws.com/mcp-memory-prod-app:latest
docker push $AWS_ACCOUNT_ID.dkr.ecr.$AWS_REGION.amazonaws.com/mcp-memory-prod-app:latest

1. Deploy infrastructure with Terraform:

cd terraform

# Initialize Terraform
terraform init

# Create terraform.tfvars
cp terraform.tfvars.example terraform.tfvars
# Edit terraform.tfvars with your values

# Plan deployment
terraform plan

# Apply infrastructure
terraform apply

1. Update ECS service (for subsequent deployments):

./scripts/deploy.sh

Connecting from Claude

Configure your Claude client with the MCP SSE endpoint:

{
  "mcpServers": {
    "mcp-memory": {
      "transport": {
        "type": "sse",
        "url": "https://your-domain.com/mcp/sse",
        "accessToken": "your-auth-token"
      }
    }
  }
}

Architecture

┌─────────────┐     ┌─────────────┐     ┌──────────────┐
│   Claude    │────▶│     ALB     │────▶│  ECS Fargate │
└─────────────┘     └─────────────┘     └──────────────┘
                           │                     │
                           │              ┌──────▼──────┐
                           │              │     RDS     │
                           │              │  PostgreSQL │
                           │              └─────────────┘
                           │                     │
                           │              ┌──────▼──────┐
                           │              │  OpenSearch │
                           │              │  Serverless │
                           │              └─────────────┘
                           │                     │
                           │              ┌──────▼──────┐
                           └──────────────│   Bedrock   │
                                         │   Titan     │
                                         └─────────────┘

Available Tools

find_person

Search for people by name, email, or organization.

get_profile

Retrieve detailed profile with attributes and recent notes.

remember_note

Add a note to an entity and index it for semantic search.

semantic_search

Search notes semantically using vector similarity.

Environment Variables

• PORT: Server port (default: 8080)
• AUTH_TOKEN: Bearer token for authentication
• PG_URL: PostgreSQL connection string
• OPENSEARCH_URL: OpenSearch endpoint
• OPENSEARCH_INDEX: Index name for vector search
• BEDROCK_REGION: AWS region for Bedrock
• AWS_PROFILE: AWS profile for local development

Monitoring

• CloudWatch Logs: /ecs/mcp-memory-prod
• ECS Console: Monitor service health and tasks
• RDS Metrics: Database performance insights
• ALB Target Health: Check target group health

Security Considerations

• All traffic encrypted with TLS
• Private subnets for compute and data
• IAM roles with least privilege
• Secrets in AWS Secrets Manager
• VPC security groups restrict access

Troubleshooting

Stream drops after ~60s

Increase ALB idle timeout in Terraform: ``hcl idle_timeout = 300 # 5 minutes ``

401 Unauthorized

Check AUTH_TOKEN in Secrets Manager matches client config.

OpenSearch returns empty

Verify index created with k-NN enabled and embeddings are being generated.

License

MIT