1. Docs
  2. Overview
  3. AI agents

AI agents

How AI agents work on the Giant Swarm platform, powered by Muster as an MCP gateway that gives assistants unified, secure access to your fleet.

The Giant Swarm platform is built to be operated by AI agents and people. You can ask an AI assistant—in your IDE or in the developer portal—questions like “are there any pods in CrashLoopBackOff on any cluster?” You get answers grounded in live cluster state, without switching between terminals and dashboards.

The piece that makes this possible is Muster, a universal control plane built on the Model Context Protocol (MCP). Muster aggregates many MCP servers—Kubernetes and Prometheus access on every management cluster, plus optionally Grafana, Teleport, and others—behind a single, OAuth-protected endpoint. Your AI assistant connects to Muster instead of juggling a separate connection and set of credentials for each server.

The pieces

  • Muster: the MCP gateway. It aggregates downstream MCP servers, presents their combined capabilities through one connection, and handles authentication on your behalf.
  • mcp-kubernetes: runs on each management cluster and exposes Kubernetes resources (pods, deployments, services, logs, events, and more) through a secure MCP API.
  • mcp-prometheus: runs alongside it on each management cluster and exposes the cluster’s metrics through MCP, so the assistant can correlate Kubernetes state with PromQL queries and alerts.
  • Your AI assistant: Claude Code, Cursor, VS Code with GitHub Copilot, the developer portal’s built-in chat, and other MCP-capable tools. It talks MCP to Muster and turns your plain-language questions into tool calls.

In this section

  • Introduction: what Muster is and the problem it solves.
  • Architecture: the aggregator design and how fleet-wide aggregation works.
  • Meta-tools: the small tool surface agents actually see, and how lazy discovery keeps context lean.
  • Security: OAuth 2.1, per-user tool visibility, and single sign-on across clusters.

To set up an assistant in your IDE today, follow Set up your AI agent.

  • Introduction to Muster
    What Muster is, the MCP-server-sprawl problem it solves, and how intelligent tool aggregation makes AI assistants on the platform cheaper and more capable.
  • Muster architecture
    How the central Muster aggregator exposes one OAuth-protected MCP endpoint, aggregates per-cluster mcp-kubernetes servers across a fleet, and why workflows cut agent token cost.
  • Meta-tools and tool discovery
    The small set of meta-tools Muster exposes to AI agents, how lazy discovery keeps the context window lean, and the prefixes that group external, core, and workflow tools.
  • AI agent security and SSO
    How Muster secures AI agent access with OAuth 2.1, per-user tool visibility keyed on identity, and single sign-on across clusters via token forwarding and exchange.