Concepts and Architecture

This chapter introduces the components, request flow, deployment models, and capabilities of the Uyuni MCP Server.

The Model Context Protocol (MCP) is an open protocol that allows an AI application to discover and call tools exposed by an external server. The Uyuni MCP Server implements MCP tools that translate structured tool calls into Uyuni HTTP API requests.

The MCP Server integrates with the existing Uyuni authorization model. The authenticated account determines which content is visible and which operations are permitted.

1. Components

An MCP integration contains the following components:

MCP host

The AI application in which the user works.

MCP client

The protocol client maintained by the host. It discovers tools, sends tool calls, displays progress and results, and can request user approval.

Uyuni MCP Server

The process that exposes tools over stdio or streamable HTTP and calls the Uyuni API.

Uyuni Server

The managed infrastructure service and authorization boundary.

Identity provider

An optional OAuth 2.0/OpenID Connect provider used for a shared HTTP deployment.

2. Request flow

For a read-only request, the flow is:

  1. The user asks the AI assistant to retrieve information from Uyuni.

  2. The AI assistant selects an MCP tool and supplies structured arguments.

  3. The MCP client sends the tool call to the MCP Server.

  4. The MCP Server authenticates to Uyuni, calls the corresponding API, and normalizes the response.

  5. The MCP client returns the structured result to the AI assistant.

State-changing tools add an approval step when the MCP client supports MCP elicitation. For more information, see Write tool controls.

3. Deployment models

Model Use case Characteristics

Client-managed stdio

One user or one local AI application

The MCP client starts one server process. Communication remains on stdin and stdout. Uyuni credentials are normally supplied to that process.

Standalone HTTP

A persistent service shared by multiple users or clients

Clients connect to /mcp over streamable HTTP. OAuth is strongly recommended. Production deployments should expose the endpoint only over TLS.

Use client-managed stdio unless the MCP Server must be shared or centrally operated. The smaller network surface and process lifetime make stdio easier to secure.

4. Capabilities and limitations

The server exposes MCP tools for systems, updates, CVEs, actions, activation keys, and system groups. Write tools are disabled by default. The server does not expose MCP resources or prompts.

Tool results can be paginated to reduce context usage. For fleet-wide questions, use summary tools first and request detailed update or CVE data only when required.

An AI-generated explanation is not an authoritative record of system state. Review the structured tool result and confirm important changes in the Uyuni Web UI, especially after scheduling updates, rebooting, bootstrapping, or removing systems.