Nomad provides a simple, flexible, and scalable alternative to Kubernetes for running diverse workloads like containers, VMs, and batch jobs.
The Model-Controller-Periphery (MCP) protocol provides a unified, open standard for AI models to interact with external tools and APIs.
LLM skills act as instruction playbooks that guide an AI's workflow, while MCP provides the action-oriented tools that execute API calls.
An MCP server for Nomad reduces context switching and cognitive load for on-call engineers by enabling cluster operations through natural language prompts.
The server implements over 50 tools covering the full Nomad API and supports multiple transport protocols like STDIO, SSE, and HTTP streaming.
Security is enforced by reusing Nomad's existing ACL token system, supplemented with an API path allowlist, origin validation, and connection timeouts.
The server benefits from separate interfaces and a single binary deployment, but faces trade-offs like increased latency with more tools and reliance on Nomad's ACLs.
The demo showcases using natural language to list jobs, drain nodes for maintenance, scale services, and troubleshoot a failed WordPress deployment.
Effective AI tool integration requires clear descriptions, a minimal set of essential tools, strict permissioning, and a simple installation process.
Jobs that call for the skills explored in this talk.