I Built an AI Agent to Reduce My AWS, GCP and Azure Bills (MCP + FastAPI)

Moiz Ali Moomin — Wed, 25 Mar 2026 18:14:36 +0000

*I gave this one prompt to an AI… and the results honestly surprised me:
*
*fetch 90 days data for aws cost and billing and give recommendation for better cost optimization
*
🚀 I Built an MCP Server That Analyzes AWS Costs Using AI (FastAPI + FinOps)

You can visit my repo the link is below:

https://github.com/Moiz-Ali-Moomin/mcp-cloud-finops-ai-agent

*Cloud bills are confusing.
*
Most of us don’t realize:
• where money is being wasted
• which resources are idle
• what to actually fix

So I built something to solve this.

👉 An MCP (Model Context Protocol) server + AI agent that analyzes AWS costs and gives actionable optimization recommendations.

*🤖 What I Built
*
I created an MCP server using FastAPI that:

• connects with AWS billing and infrastructure data
• allows an AI agent to analyze usage
• returns real optimization insights
• provides exact CLI commands to fix issues

*⚡ How It Works
*

MCP server exposes tools (billing + infra data)
AI agent calls these tools
Data is analyzed (90 days usage)
System generates: • cost insights • anomalies • optimization suggestions

*🧪 The Prompt
*
I gave a simple instruction:

“Fetch 90 days AWS cost and suggest optimizations”

*📊 What the System Returned
*
• Total spend: $11.44
• Top cost driver: EC2-Other
• Estimated savings: ~40%

👉 Almost half the cost could be optimized.

*🔍 Key Insights (Automatically Generated)
*

AWS Cost Explorer was costing money
• I was paying just to query billing data
• Suggested fix: use S3 Cost & Usage Reports + caching
Unused resources detected
• orphaned EBS volumes
• unused Elastic IPs
• idle infra

Suggested command:
aws ec2 describe-volumes --filters Name=status,Values=available

Idle NAT Gateway • charging even without usage

Suggested fix:
terraform destroy

Cost spikes explained • Jan spike → ECS usage • March spike → Kubernetes testing

👉 AI correlated infra activity with billing automatically

*🧠 Why MCP Makes This Powerful
*
Instead of hardcoding logic:

• MCP enables tool-based architecture
• AI dynamically decides what to fetch and analyze
• system becomes flexible and extensible

💻 Example Output

{
"resource": "EC2",
"issue": "Underutilized",
"recommendation": "Downgrade instance",
"estimated_savings": "30%"
}

*🧪 Try It Yourself
*
GitHub repo:
https://github.com/Moiz-Ali-Moomin/mcp-cloud-finops-ai-agent/

Steps:

git clone https://github.com/Moiz-Ali-Moomin/mcp-cloud-finops-ai-agent
pip install -r requirements.txt
uvicorn main:app --reload

*🤝 Contributing
*
If this sounds interesting, contributions are welcome.

You can help by:

• adding support for more cloud services
• improving cost analysis logic
• building a frontend/dashboard
• adding auto-remediation features

Feel free to:
• open issues
• suggest improvements
• submit pull requests

👉 Even small contributions are appreciated.

*🚀 What’s Next
*
• real-time monitoring
• multi-cloud support
• auto-remediation (AI fixing issues automatically)

*💬 Final Thoughts
*
Before this, cloud cost optimization meant:
• dashboards
• manual effort
• guesswork

Now it’s:

👉 Ask AI → get exact fixes

If you're into DevOps, cloud, or AI agents, would love your feedback.

DEV Community: Moiz Ali Moomin

Boost

I Built an AI Agent to Reduce My AWS, GCP and Azure Bills (MCP + FastAPI)

I Built an AI Agent to Reduce My AWS, GCP and Azure Bills (MCP + FastAPI)