aws-cost-optimizer

name	aws-cost-optimizer
description	Analyze AWS costs, find waste, and recommend optimizations using read-only MCP tools
version	1.0.0
author	kai-agent
metadata	{"kai":{"tags":["kai","aws","cost","optimization","finops"],"related_skills":["aws-performance"]}}

AWS Cost Optimizer

Analyze AWS spending, identify waste, and recommend cost optimizations. All operations are read-only — no changes are made to infrastructure.

Prerequisites

• Workspace with AWS credentials configured in kai-backend
• MCP tools available: mcp__kai__aws_*

Workflow 1: Quick Cost Check

Get a snapshot of current spending and trend.

1. Get last 30 days of costs:

   aws_cost_summary(workspaceId, startDate="YYYY-MM-DD", endDate="YYYY-MM-DD")
   

   • Identify top 5 spending services
   • Note total monthly spend

2. Forecast next 30 days:

   aws_cost_forecast(workspaceId, startDate="today", endDate="+30d", granularity="MONTHLY")
   

   • Compare with current spend
   • Flag if forecast shows >10% increase

3. Present findings as a table:

   Service	Last 30d	Forecast	Trend

Workflow 2: Find Waste

Systematically identify unused/idle resources.

1. Unused EBS volumes (easy win):

   aws_ec2_unused_volumes(workspaceId)
   

   Each unattached volume costs money. Calculate: size_gb * $0.10/month

2. Unattached Elastic IPs:

   aws_ec2_unattached_eips(workspaceId)
   

   Each costs ~$3.65/month

3. Idle EC2 instances (biggest savings potential):

   aws_ec2_list_instances(workspaceId, state="running")
   

   Then for each instance, check CPU over 7 days:

   aws_cloudwatch_get_metrics(
     workspaceId, namespace="AWS/EC2", metricName="CPUUtilization",
     dimensions=[{"name":"InstanceId","value":"i-xxx"}],
     startTime="-7d", endTime="now", period=3600, statistics=["Average"]
   )
   

   • Flag instances with avg CPU < 10% as "idle — consider downsizing or stopping"
   • Flag instances with avg CPU < 2% as "likely unused — consider terminating"

4. Oversized Lambda functions:

   aws_lambda_list_functions(workspaceId)
   

   Then check actual duration vs configured memory:

   aws_cloudwatch_get_metrics(
     workspaceId, namespace="AWS/Lambda", metricName="Duration",
     dimensions=[{"name":"FunctionName","value":"xxx"}],
     startTime="-7d", endTime="now", period=86400, statistics=["Average","Maximum"]
   )
   

   • If avg duration < 100ms but memory > 512MB, suggest reducing memory

Workflow 3: Coverage Analysis

Find gaps in Reserved Instances / Savings Plans.

1. RI coverage:

   aws_reservation_coverage(workspaceId, startDate="-30d", endDate="today")
   

   • Identify services running >80% on-demand
   • Calculate potential savings with 1-year no-upfront RI

2. Savings Plans coverage:

   aws_savings_plans_coverage(workspaceId, startDate="-30d", endDate="today")
   

   • Quantify uncovered spend

Workflow 4: S3 Cost Optimization

1. List all buckets:

   aws_s3_list_buckets(workspaceId)
   

2. Analyze each significant bucket:

   aws_s3_bucket_analysis(workspaceId, bucketName="xxx")
   

   • Flag buckets without lifecycle rules (data never expires)
   • Flag buckets without versioning cleanup (old versions accumulate)
   • Recommend Intelligent-Tiering for infrequently accessed data

Reporting Format

Present findings as:

Cost Optimization Report

Current monthly spend: $X,XXX Estimated monthly savings: $XXX (X%)

Finding	Service	Monthly Cost	Savings	Severity
5 unused EBS volumes	EC2	$50	$50	Low
3 idle instances (CPU <5%)	EC2	$450	$300	High
No lifecycle rules on logs bucket	S3	$200	$150	Medium

Recommendations (by impact):

1. ...
2. ...

Rules

• Always be specific with dollar amounts — use actual data, not estimates
• Never recommend changes without showing the supporting metrics
• Group findings by severity: Critical (>$500/mo), High ($100-500), Medium ($20-100), Low (<$20)
• If a resource has tags suggesting it's intentionally running (e.g., "environment=production"), note that before recommending termination