Ansible MCP Server Live Demo: AI-Driven Automation in Cursor IDE

At Red Hat Tech Day Netherlands (June 2026), Fred van Zeile demonstrated the Ansible MCP Server live — connecting Cursor IDE to a real Ansible Automation Platform instance and running agentic automation workflows through natural language.

This isn’t a slide deck or a video. This is real-time AI driving actual infrastructure automation.

The Setup: 6 MCP Servers, 107 Tools

The demo used Cursor IDE (Enterprise Plan) with six MCP server connections:

Total: 107 tools exposed to the AI agent through a standardized MCP interface.

The Configuration: mcp.json

The configuration is straightforward JSON:

{
  "mcpServers": {
    "aap-mcp-job-agent": {
      "type": "http",
      "url": "https://aap.fvz.ansible-labs.de:8448/job_management/mcp",
      "headers": {
        "Authorization": "Bearer <token>"
      }
    },
    "aap-mcp-inventory-agent": {
      "type": "http",
      "url": "https://aap.fvz.ansible-labs.de:8448/inventory_management/mcp",
      "headers": {
        "Authorization": "Bearer <token>"
      }
    },
    "aap-mcp-system-mon": {
      "type": "http",
      "url": "https://aap.fvz.ansible-labs.de:8448/system_monitoring/mcp",
      "headers": {
        "Authorization": "Bearer <token>"
      }
    },
    "aap-mcp-user-agent": {
      "type": "http",
      "url": "https://aap.fvz.ansible-labs.de:8448/user_management/mcp",
      "headers": {
        "Authorization": "Bearer <token>"
      }
    },
    "aap-mcp-sec-comp": {
      "type": "http",
      "url": "https://aap.fvz.ansible-labs.de:8448/security_compliance/mcp"
    }
  }
}

Key detail: it’s just HTTP endpoints with Bearer token authentication. Any MCP-compatible client (Cursor, Claude Desktop, VS Code, custom apps) can connect.

Demo 1: “Show Me the Last 5 Jobs from AAP”

The agent queries AAP and returns a structured table:

Notes from the agent:

• All five ran against Demo Inventory on project Demo Project
• Jobs 76 and 77 used the Paint job template; the rest used Demo Job Template
• Extra vars included color: Yellow (Job 76), Green (Job 77), and Rood (Jobs 73-75)

Demo 2: “Run the Paint Job with Color Purple”

The agent attempts to launch the Paint job template via MCP:

Checking how to launch the Paint job template via AAP MCP.

⚡ Ran Job Templates Launch Create in aap-mcp-job-mgmnt

The Paint job template (ID 43) rejected the launch with this validation error:

Value 'Purple' for color expected to be one of ['Green', 'Yellow', 'Red'].

The template’s survey only allows those three colors, so Purple cannot be used without updating the job template survey first.

Would you like me to:

1. Run it with one of the allowed colors (Green, Yellow, or Red), or
2. Help update the Paint job template to add Purple as an option?

This is the key safety feature — the MCP server respects AAP’s survey validation. The AI can’t bypass guardrails.

Demo 2b: “Ok, Just Use Green”

The Paint job has been launched and completed successfully.

Field          Value
Job ID         78
Name           Paint
Template       Green
Color          ✅ successful
Status         successful
Playbook       hello_world.yml
Inventory      Demo Inventory
Duration       5.3s
Hosts          1 ok

From natural language request to successful job execution in seconds.

Demo 3: “Show Me the Last Failed Job and Analyse What Went Wrong”

This is where it gets powerful. The agent performs multi-step investigation:

The investigation flow:

1. Thought briefly — Finding the most recent failed job
2. Ran Jobs List in aap-mcp-job-mgmnt
3. Ran Jobs Retrieve in aap-mcp-job-mgmnt
4. Ran Jobs Stdout Retrieve in aap-mcp-job-mgmnt
5. Ran Jobs Job Events List in aap-mcp-job-mgmnt
6. Read jobs_list.json (L1-21), jobs_stdout_retrieve.json (L1-27), jobs_job_events_list.json

The Root Cause Analysis

Root cause analysis

Two issues line up:

1. Wrong VPC name: caap instead of aap

The failed workflow used: "ec2_vpc_name": "caap"

A later successful rerun of the same workflow (jobs #24-26) used: "ec2_vpc_name": "aap"

caap looks like a typo for aap. That likely caused the playbook to create or look up the wrong VPC stack.

2. Parallel jobs racing on the same VPC name

Workflow #18 launched two EC2 deploy jobs at once (prometheus + node1), both with ec2_vpc_name: caap. Both tried to create VPC/subnet/IGW/route-table resources with the same tags at the same time. That can leave duplicate or ambiguous route tables in AWS — exactly what triggers “Tags provided do not identify a unique route table”.

How to fix it

Immediate fix (most likely sufficient):

1. Update the workflow Deploy EDA Prometheus Demo 1 so ec2_vpc_name is aap, not caap
2. Relaunch the workflow.

The agent identified a typo (caap vs aap) AND a race condition from parallel job execution — all by reading job events and stdout through MCP tools.

Demo: Failed Job Summary

The failed job details:

• Job ID: 20
• Name: EDA - Deploy Instance on EC2
• Template: EDA - Deploy Instance on EC2 (ID 34)
• Workflow: Deploy EDA Prometheus Demo 1 (#18)
• Playbook: instance.yml
• Inventory: ec2
• Status: failed
• Duration: 13.2s
• Extra vars: ec2_vpc_name: caap, instance_name: node1

The most recent failed job is Job #20 (Mar 17, 2026). Job #10 failed at the same time with the same root cause — both were launched in parallel by workflow Deploy EDA Prometheus Demo 1 (workflow job #18).

Key Takeaways

1. 107 tools across 6 domains — comprehensive AAP coverage through MCP
2. Survey validation enforced — AI can’t bypass AAP guardrails
3. Multi-step root cause analysis — agent autonomously queries jobs, stdout, events
4. Natural language → real automation — no API knowledge required
5. Any MCP client works — Cursor, Claude Desktop, custom apps
6. Standard HTTP + Bearer auth — simple, secure configuration

The Bigger Picture

This demo shows what “AI-driven automation” actually means in practice:

• Not replacing Ansible playbooks
• Not generating code and hoping it works
• Actually using AI as an intelligent operator that understands platform constraints, investigates failures, and executes within guardrails

The MCP server is the bridge between natural language intent and deterministic automation execution.

Related Content

• Ansible AIA + MCP Server + BYOK RAG
• AAP 2.7 Automation Portal and EE Builder
• Ansible VS Code Extension with MCP Server
• Upgrading to Ansible Automation Platform 2.6+