AI Agents Just Infiltrated CloudOps
AWS Labs just dropped an MCP server for the AWS CLI. AI Agents just infiltrated CloudOps.
This is huge. CloudOps just got way more productive. And if that describes your role, you just got a new skillset you need to add to your quiver right now.
This is exactly what I was talking about in my recent post about intelligence-over-infrastructure, if in more a more technical context. Now you can chat with your AWS environment, which is deceptively powerful.
For instance, you can now type in this single prompt:
Show me every EBS volume larger than 500GB that isn’t attached to anything, older than 30 days, and tell me what it would cost to store them for another month.
Which would previously have required you to write this bash script:
# Pull every volume in the region
aws ec2 describe-volumes --region $AWS_REGION --output json > /tmp/vols.json
# Keep only: size >500 GiB, no Attachments, older than 30 days
cutoff=$(date -u -d '30 days ago' +%s)
jq -r --arg cutoff "$cutoff" '
.Volumes[]
| select(.Size>500
and (.Attachments|length==0)
and ((.CreateTime|fromdateiso8601)<($cutoff|tonumber)))
| [.VolumeId,.Size,.VolumeType,.AvailabilityZone]
| @tsv' /tmp/vols.json > /tmp/candidates.tsv
# Look up the $/GiB-month price for the volume type.
# Example shown for gp3 in us-east-1; adjust volumeApiName/location as needed.
price=$(aws pricing get-products \
--service-code AmazonEC2 \
--region us-east-1 \
--filters 'Type=TERM_MATCH,Field=volumeApiName,Value=gp3' \
'Type=TERM_MATCH,Field=location,Value="US East (N. Virginia)"' \
--query 'PriceList[0]' --output text | \
jq -r '.terms.OnDemand[][].priceDimensions[]?.pricePerUnit.USD')
# Report projected storage cost for another month
awk -v p="$price" '{printf "%s\t%s GiB\t$%.2f/month\n",$1,$2,$2*p}' /tmp/candidates.tsv
I’m just at the beginning of thinking about the implications of this and playing with it in my own AWS account. Some initial thoughts on where such an MCP server might really shine:
- Self-healing guardrails: Agents that not only detect misconfigurations but fix them (after approval).
- Cross-account governance: Iterate through every member account in an AWS Organization, aggregate findings, and patch issues.
- FinOps automation: Agent that periodically identifies orphaned or idle resources, estimates waste, and deletes them (after approval).
- Q&D sandboxes: Spin up entire ephemeral stacks from chat, run experiments, and tear them down without ever opening the Console (which I personally find not especially user-friendly).
- AI Orchestration: Combine with other MCP servers to orchestrate complex actions without needing to hard-code CLIs or APIs.
Bottom line: CloudOps just got way more powerful and productive, and the field will be split into people who know how to use this new power and those who don’t. Guess which ones will be lookoing for jobs after the next RIF.