Overview

The Red Hat Insights MCP Server (formerly Insights MCP; part of Red Hat Lightspeed) is a lightweight, self-hosted Model Context Protocol (MCP) server that links MCP-compatible LLM agents to Red Hat Lightspeed services. It exposes read-oriented APIs for inventory, vulnerability scanning, advisor recommendations, image/build services and other Lightspeed toolsets so agents can ask natural language questions and receive structured results from Red Hat services.

This server is designed for development, automation, and secure experimentation with LLM-based assistants. It runs in read-only mode by default (minimizing risk), supports service-account or JWT authentication for Lightspeed APIs, and can be configured to allow write operations only when explicitly enabled and authorized via RBAC.

Repository: https://github.com/RedHatInsights/insights-mcp

Features

Read-first MCP server with optional write tools (enable with –all-tools)
Connects LLM agents (MCP-compatible clients like VSCode extensions, Cursor, etc.) to Red Hat Lightspeed services
Natural-language query capability for Lightspeed data (inventory, vulnerabilities, advisor, planning, remediations)
Service-account and JWT authentication support for Lightspeed APIs
RBAC-aware: actions the server can perform are limited by the service account’s roles
Small footprint and self-hostable via Podman/Docker

Supported Lightspeed Services

Service	Purpose
advisor	Configuration and recommendations for RHEL systems
hosted image builder	Build and manage OS images (hosted image builder)
inventory	Host and system inventory information
planning	Roadmap/planning data for system lifecycle
remediations	Remediation workflows and patches
vulnerability	CVE and vulnerability assessments for hosts and images

Installation / Configuration

Prerequisites: podman or docker. Create a Red Hat service account (Client ID + Client Secret) with the minimal roles required by the tools you want to use.

Example environment variables:

# Required for service-account authentication
LIGHTSPEED_CLIENT_ID=xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx
LIGHTSPEED_CLIENT_SECRET=your-client-secret
# Optional: override server listen port
PORT=8080

Run with Podman:

podman run -d --name lightspeed-mcp \
  -e LIGHTSPEED_CLIENT_ID="$LIGHTSPEED_CLIENT_ID" \
  -e LIGHTSPEED_CLIENT_SECRET="$LIGHTSPEED_CLIENT_SECRET" \
  -p 8080:8080 \
  quay.io/redhatinsights/insights-mcp:latest

Run with Docker:

docker run -d --name lightspeed-mcp \
  -e LIGHTSPEED_CLIENT_ID="$LIGHTSPEED_CLIENT_ID" \
  -e LIGHTSPEED_CLIENT_SECRET="$LIGHTSPEED_CLIENT_SECRET" \
  -p 8080:8080 \
  quay.io/redhatinsights/insights-mcp:latest

Start with write tools enabled (explicit):

podman run -d --name lightspeed-mcp \
  -e LIGHTSPEED_CLIENT_ID="$LIGHTSPEED_CLIENT_ID" \
  -e LIGHTSPEED_CLIENT_SECRET="$LIGHTSPEED_CLIENT_SECRET" \
  -e MCP_TOOL_OPTIONS="--all-tools" \
  -p 8080:8080 \
  quay.io/redhatinsights/insights-mcp:latest

JWT bearer token mode (client provides token to MCP server via SSE/HTTP):

Do not set LIGHTSPEED_CLIENT_ID / SECRET in the server.
Client will include Authorization: Bearer <token> in SSE/HTTP messages.

Example mcp.json entry for an LLM client:

{
  "name": "lightspeed-mcp",
  "url": "http://mcp.example.local:8080",
  "transport": "sse"
}

Available Tools / Resources

The MCP server provides a set of toolsets (see toolsets.md in the repository) exposing operations like:

advisor tools: fetch configuration issues and recommendations
inventory tools: list hosts and meta-data
vulnerability tools: scan host/image vulnerabilities and CVE details
hosted image builder tools: view/build image artifacts (write operations gated)
remediations tools: read or create remediation playbooks (write gated)
planning tools: read planning/roadmap data

Check the repository’s toolsets.md for the complete list and API shapes.

Use Cases

Scan container or VM images for known CVEs and produce summaries for an LLM-based assistant to explain impact and fixes.
- Agent: “List high severity CVEs affecting image X and suggest remediation steps.”
- MCP: queries vulnerability and inventory services, returns structured vulnerabilities + package info.
Produce targeted recommendations for a fleet of hosts.
- Agent: “Show top 5 advisor recommendations for RHEL hosts in group A.”
- MCP: uses advisor + inventory toolsets to deliver prioritized results.
Integrate into developer workflows to inspect image build status or manifest details.
- Agent: “Did the latest compose for image builder complete? Show artifacts.”
- MCP: queries hosted image builder service; if write enabled, can create or trigger builds.
Safe experimentation with LLMs: run the server locally in read-only mode to limit accidental writes while working with agents.

Permissions and Security

Service accounts start with no roles. An organization admin must add the account to a group that grants appropriate roles (e.g., “Vulnerability viewer”, “Inventory Hosts viewer”, “RHEL Advisor viewer”, “Remediations user”).
Default server mode is read-only. Only enable write capabilities when you understand and have assigned the required RBAC roles.
Do not expose a server with client secrets publicly. If connecting remote clients, prefer JWTs and secure transport.

Emergency revocation (“kill switch”):

Stop the MCP process/container immediately (e.g., podman stop <container_id>).
In the Red Hat Console, reset or delete the service account credentials (Console → Service Accounts → Delete/Reset).
Remove the MCP entry from client configurations (e.g., mcp.json) to prevent automatic reconnection.

Red Hat Insights MCP Server: Images, Vulnerabilities, Recommendations