Deployment Roadmap
Aegis MCP Gateway is designed to support different environments depending on requirements. Single-tenant development workloads run in a single process, while enterprise production environments scale across high-availability cloud infrastructure.
Because the core engine is decoupled from underlying datastores (using the Ports & Adapters design), migrating from a Docker Compose evaluation (P1) to a full multi-node Kubernetes cluster (P2) or a managed SaaS footprint (P3) requires only configuration edits—not code rewrites.
Deployment Options
Phase 1 — Single-Node Compose Sandbox
Ideal for local developers, testing loops, and offline, air-gapped pilots. The entire control plane and data plane co-exist in a single process. All datastores are run inside local containers:
- State Storage: Single-instance PostgreSQL database using EF Core migrations.
- Cache & Rate Limiting: In-process memory cache or single-node local Redis.
- Audit Logs: Exported directly to the PostgreSQL database with Full-Text Search.
- Secrets: Retrieved from local environment variables or basic configuration files.
# Run from source in local developer mode
dotnet run --project src/Mcp.Gateway.Host --role=all
Phase 2 — Kubernetes & Helm HA Cluster
Designed for production loads inside a corporate VPC. In this phase, Control Plane and Data Plane pods are separated into distinct deployments, allowing them to scale independently:
- Data Plane: Horizontally scaled via Horizontal Pod Autoscaling (HPA) using CPU and connection metrics.
- State & Cache: Connects to high-availability database clusters (e.g. Postgres-HA and Redis Sentinel).
- Audit Logs: Written asynchronously to OpenSearch for query indexing and archived to MinIO using WORM (Write Once, Read Many) policies.
- Secrets: Integrates with HashiCorp Vault.
# Deploy Aegis high-availability cluster using Helm
helm repo add aegis https://charts.aegis-gateway.org
helm install mcp-gateway aegis/mcp-gateway -f prod-values.yaml
Phase 3 — Managed SaaS (AWS)
Our fully managed multi-tenant SaaS offering. Data plane instances run in EKS clusters across multiple regions. The gateway integrates natively with cloud resources:
- State Storage: Amazon Aurora PostgreSQL (Multi-AZ with regional replica nodes).
- Caching: Amazon ElastiCache for Redis (HA clusters).
- Event Bus: Amazon SNS and SQS for async message queues and event processing.
- Audit Archives: Amazon S3 buckets configured with Object Lock (WORM) for permanent audit compliance.
- Identity & Secrets: Amazon Cognito for IdP federation, and AWS Secrets Manager for credentials.
Multi-Tenancy Isolation Modes
Aegis handles multi-tenancy at the architectural level. Every database entity, query, and cache key includes a tenant_id. We support three different isolation modes depending on compliance requirements:
| Isolation Tier | Infrastructure Strategy | Best For | Details |
|---|---|---|---|
| Logical (RLS) | Shared database; tenant data isolated via Row-Level Security (RLS) policies. | Standard SaaS users | Low cost, high density. Redis keys are namespaced per tenant. |
| Strong Logical | Dedicated schema or database per tenant. | Enterprise SaaS users | Prevents data leakage. Isolates compute costs per database connection. |
| Physical | Dedicated cluster deployed in a separate VPC or network. | Regulated enterprise buyers | Absolute physical isolation of data planes and control planes. |