QCon CALM Demos: An End-to-End Story

🔴 Difficulty: Advanced | ⏱️ Time: ~30–60 minutes for the full flow

Overview

A scripted, five-scenario walkthrough of several CALM-based scenarios, originally presented at the QCon Conference. Together the scenarios trace the lifecycle of a real platform, from initial deployment, through control and governance, to fully autonomous agents, with every step grounded in a CALM architecture.

Run the scenarios in sequence (each builds on state left by the previous one) using the orchestrator script.

The Five Scenarios

#	Scenario	What it Demonstrates
1	Deploy API & MCP Architecture	Generate Kubernetes manifests from a CALM architecture using calm template, then deploy a Trades REST API, an MCP server, and an A2A server onto a Minikube cluster.
2	Introducing Controls and Governance	Add a declarative MCP guardrail control to the architecture, regenerate, and watch CALM controls turn into Kubernetes ConfigMaps that the running MCP server picks up.
3	Gating Deployments	Block deployments unless the architecture references an approved pattern registered in CALM Hub and conforms to that pattern's required controls.
4	Scaling Deployments and Operational Change	Roll out a platform-wide change (resource limits on every workload) by releasing a new template bundle version, without touching a single team-owned architecture file.
5	Rapid Platform Adoption	Spin up an autonomous rebalancer agent that communicates via the Agent-to-Agent (A2A) protocol — OBSERVE → DECIDE → ACT, no human in the loop.

Concepts at Work

Concept	Where It Shows Up
calm template — architecture-to-infrastructure generation	All scenarios
CALM Patterns — reusable, validated architecture templates	Scenarios 3, 4
CALM Controls — declarative governance requirements	Scenarios 2, 3
Template bundles — platform-owned generation logic, versioned independently from architectures	Throughout (foregrounded in scenario 4)
CALM Hub — registry for patterns, namespaces, and entitlements	Scenario 3
calm validate — architecture conformance to its pattern	Scenario 3
MCP (Model Context Protocol)	Scenarios 1, 2
A2A (Agent-to-Agent) protocol	Scenario 5
Kubernetes manifests, network policies, ConfigMaps	Scenarios 1–4

Prerequisites

• minikube — local Kubernetes cluster
• kubectl — cluster management
• docker + docker-compose — for CALM Hub (scenario 3) and the Agent UI (scenario 5)
• calm CLI — npm install -g @finos/calm-cli
• jq — JSON processing
• bat, tree — optional, nicer terminal display (each falls back gracefully if missing)
• Node.js 22+ for the CLI itself

Verify the CLI is ready:

calm --version   # 1.46.0 or later

Kick Off the Demo

Clone the architecture-as-code repository and run the orchestrator script from the repo root:

cd conferences/qcon
./demo-flow.sh

You'll be asked to choose a mode at startup:

• Concise mode (default) — every command is shown, narration is trimmed. Good for live presentations.
• Story mode — fuller narration explaining why each step matters. Good for self-paced learning.

The orchestrator demo script walks you through scenarios 1 through 5 with Press Enter prompts between sections. If a step fails (for example, CALM Hub isn't reachable when scenario 3 starts), you'll be offered the option to retry, auto-start the missing service via Docker Compose, or quit.

Port-Forwarding for Scenarios 2–5

After scenario 1 deploys the services to Minikube, open a second terminal and start port-forwarding so the later scenarios can reach the cluster:

cd conferences/qcon/scenario1
./port-forward.sh

Services are then available at:

• MCP Server — http://localhost:8080
• Trades API — http://localhost:8081
• A2A Server — http://localhost:9103

Leave that terminal running until the demo is finished.

Connecting Claude to the MCP Server (Optional)

Once scenario 1 or 2 is running with port-forwarding active:

1. In a separate terminal, expose the MCP server with ngrok http 8080
2. Copy the ngrok HTTPS URL
3. Configure Claude to use that URL as its MCP server endpoint

After scenario 2, the guardrail is enforced: queries for restricted symbols (VOD, GME, AMC) are rejected.

Cleanup

# stop Kubernetes
kubectl delete -k conferences/qcon/scenario4/generated/kubernetes
minikube stop --profile secure
minikube delete --profile secure

# stop CALM Hub
cd calm-hub/deploy-qcon && docker-compose down

# stop the Agent UI
docker rm -f qcon-agent-ui

Learn More

• README in conferences/qcon/ — the source-of-truth doc this page is derived from, plus per-scenario READMEs
• CALM Patterns — the building block scenario 3 leans on
• CALM Controls — the governance primitive used in scenarios 2 and 3
• Model Context Protocol
• Agent-to-Agent Protocol