Skip to main content
Version: Next

MCP Hosting

The MCP Hosting layer runs and manages MCP servers directly within Fronteir AI. It handles deployment, lifecycle management, and runtime isolation for MCP servers.

Runtime Types​

Server Types​

  • Single-user: Each user gets their own isolated instance with separate credentials
  • Multi-user: A shared instance serves multiple users with shared or per-user credentials
  • Remote: External MCP servers accessed via HTTP, not hosted by Fronteir AI
  • Composite: Combines multiple servers into a single virtual server with curated tools

Deployment Environments​

Docker​

When running Fronteir AI with Docker, MCP servers are deployed as sibling containers:

  • Fronteir AI communicates with the Docker daemon to manage containers
  • Servers run alongside the Fronteir AI container
  • Suitable for development and small deployments
  • See Docker Deployment for setup details

Kubernetes​

For production deployments, Fronteir AI can deploy MCP servers to Kubernetes:

  • Servers run as pods in the cluster
  • Supports resource limits, network policies, and scaling
  • See MCP Deployments in Kubernetes for configuration details

Authentication​

Fronteir AI handles OAuth 2.1 flows for MCP servers that require authentication:

  • OAuth credentials stored securely with encryption at rest
  • Automatic token refresh
  • Per-user credential isolation
  • Supports custom OAuth configurations

See MCP Server OAuth Configuration for details on configuring OAuth for MCP servers.

Learn More​