Demo: MCP Servers — Agentic Automation with Governance
Start here. A customer watches an AI assistant query a live OpenShift cluster and launch an Ansible job template — reading everything, changing nothing except through the governed automation path — in about twenty minutes.
| Length | 20 minutes (15 + 5 for questions) |
| Audience | Platform engineers and automation leads evaluating how AI fits into operations |
| Reader | The Ansible pre-sales engineer presenting it |
| Needs a live environment? | Yes — the compelling beats are live queries against a real cluster |
| Status | Draft — OpenShift MCP content complete, AAP MCP content is placeholder |
The five documents (plus three)
| File | Read it when |
|---|---|
run-sheet.md |
While presenting. Minute markers, what is on screen, exact commands, recovery moves |
talk-track.md |
While rehearsing. The narrative and the actual words, beat by beat |
architecture.md |
When asked "how does that work". The seven servers, the credential flow, the access posture |
objections.md |
Before you go in. What this audience asks — especially the security questions |
server-inventory.md |
When asked "what can it do". Full tool listings, status tables, verification commands |
building-a-server.md |
When asked "what about my platform". When no MCP server exists: whether to build one, what language, transport, packaging, hosting |
servicenow.md |
When asked about ITSM. Why the native console is the one to wait for, and why the Ansible write path needs no MCP server |
Present from the run sheet. Rehearse from the talk track. The other five are reference.
Why the extra files. The detailed status tables, per-server tool listings,
and verification commands that run-sheet.md and architecture.md cite would
overwhelm either one — the same reason Private Automation Hub earned
clickops.md. The last two answer the
question this demo reliably provokes — "what about the platform I care about,
which has no MCP server?" — which is a different question from how the six
servers here work.
The 60-second version
- Show the six MCP servers and their access posture — three OpenShift (local; sandbox and edge read-write, demo read-only), two AAP (in-cluster, the same sandbox/demo split), one Grafana Cloud (local, read-only Viewer). The environment is in the server's name, and the audience sees the posture before a single query runs.
- Issue a read-only query against the cluster — ask what VMs are running, what pods are in a namespace, what a guest's OS version is. The assistant gets real data, not a summary or a slide.
- Launch a job template through AAP — the governed write path. The assistant requests a change, AAP enforces its own RBAC and survey, the audit trail is in the controller.
What the demo is actually about is not that AI can talk to infrastructure. It is that the boundary between what the AI can read and what it can change is visible, enforced, and auditable — and that the change path is the same governed Ansible workflow the customer already trusts.
Why it mostly works without a cluster
The status tables in server-inventory.md are committed
and render on GitHub. The architecture diagram is Mermaid. The tool listings
were measured on live servers and documented with their provenance.
But the demo's strongest beats need a warm environment. A live query returning real cluster state is the cold open, and no screenshot replaces it. If you are presenting without a cluster, lead with the architecture and the status tables, and be upfront that these are documented measurements rather than a live run.
If you want to run it live
/sales-demos-mcp # sets up all six servers — kubeconfigs + AAP tokens + Grafana
Then verify the servers answer — the skill does this automatically, but if you want to confirm independently:
claude mcp list # all six servers should appear
New to this repo? Run /sales-demos-first-time first.
Red Hat resources
| Resource | What it covers |
|---|---|
| Model Context Protocol specification | The protocol itself — what MCP is and why it exists |
| kubernetes-mcp-server | The upstream project powering the OpenShift MCP servers (Apache-2.0) |
| Deploying the Ansible MCP Server | Official Red Hat guide for the AAP MCP server |
| Deploy MCP servers on OpenShift using ToolHive | Red Hat Developer article on containerised MCP hosting |
| ansible.mcp_builder | Red Hat CoP collection for building MCP servers into container images |
Related
../../plan/platform-addons-plan.md— why the MCP servers are built this way: the transport decision, the local-first rationale, the in-cluster AAP deployment../../plan/network-mcp-plan.md— the next use case: network vendor MCP serversROADMAP.md— what is done and what is notsales.demos— the automation itself