AWS Cost Saver
Find what's wasting money in your AWS account.
You: "Scan my AWS for cost savings"
Claude: Found 8 issues. Potential savings: $340/month
Installation
The plugin ships with both a .claude-plugin/plugin.json (Claude Code) and a .codex-plugin/plugin.json (Codex) manifest, so the same source tree installs cleanly into either agent.
Claude Code
/plugin marketplace add prajapatimehul/claude-aws-cost-saver
/plugin install aws-cost-saver@aws-cost-saver-marketplace
Or interactively: /plugin → Marketplaces → Add Marketplace → prajapatimehul/claude-aws-cost-saver.
Codex
codex plugin marketplace add prajapatimehul/claude-aws-cost-saver
Then launch Codex and run /plugins to browse and install aws-cost-saver.
Other agents (MCP only)
Any agent that supports MCP can wire the AWS API server directly using plugins/aws-cost-saver/.mcp.json as a template. Skills live under plugins/aws-cost-saver/skills/ and can be loaded as plain Markdown.
Quick Start
/aws-cost-saver:scan
Or just ask: Scan my AWS account for cost savings
Requirements: AWS credentials configured (aws configure or SSO), uv installed.
This tool only reads data — it never modifies or deletes anything. A PreToolUse hook enforces a read-only allowlist: only describe-/get-/list-/lookup-/search- style AWS CLI operations are permitted, and everything else is denied by default.
Features
- 180 checks across EC2, RDS, S3, Lambda, ECS, EKS, Aurora, SageMaker, and 30+ services
- Parallel scanning - 11 domain agents run simultaneously
- Confidence scoring - filters false positives
- Real pricing - uses AWS Cost Explorer
- Markdown reports - clean, actionable output
What's included
Commands (slash)
| Command | Description |
|---|---|
/aws-cost-saver:scan | Full cost optimization scan across 11 domains |
Skills (auto-loaded)
| Skill | Description |
|---|---|
reviewing-findings | Confidence-scored review; filters false positives |
validating-aws-pricing | Mandatory Pricing-API/verified-table sanity check before reporting |
Subagent
aws-cost-saver:aws-cost-saver — domain-scoped scanner. Invoke 11 in parallel (one per domain) for a full account audit.
MCP server
The plugin pre-wires the AWS API MCP server in .mcp.json, launched via uvx from scripts/start-mcp.sh. A PreToolUse hook in hooks/hooks.json (backed by scripts/guard_readonly.py) denies every non-read AWS CLI operation, keeping every scan read-only by default.
CLI
The repository also ships a standalone Python CLI (main.py) usable without a coding agent:
python main.py checks # list all 180 checks
python main.py check EC2-001 # detail for one check
python main.py scan-info # show the AWS CLI commands a scan runs
python main.py spend-hotspots --profile X # rank scan order by real spend (Cost Explorer)
python main.py report --findings findings.json
Demo
Results

Before & After
Real AWS account: $105/day → $42/day after running the scanner
How it works
Step 1: Choose compliance requirements

Step 2: Parallel agents scan your account

Domains & Checks
| Domain | Checks | Key Areas |
|---|---|---|
| Compute | 28 | EC2 idle/over-provisioned, Compute Optimizer ML, Cost Optimization Hub, EBS, GP2→GP3 |
| Storage | 24 | S3 lifecycle, CloudWatch Logs, snapshots, Secrets Manager |
| Database | 17 | RDS idle/over-provisioned, extended support, Valkey migration, RI coverage |
| Networking | 19 | Unused EIPs, NAT, public IPv4, data transfer, VPC endpoints |
| Serverless | 11 | Lambda memory, unused functions, SQS provisioned-mode idle |
| Reservations | 13 | RI/SP/Database SP coverage, purchase recommendations |
| Containers | 17 | ECS/EKS idle, extended support versions, Fargate, Spot, ECR lifecycle |
| Advanced DBs | 18 | Aurora, DocumentDB, Neptune, Redshift |
| Analytics | 15 | SageMaker, EMR, OpenSearch |
| Data Pipelines | 12 | Kinesis, MSK, Glue |
| Storage Advanced | 6 | FSx, AWS Backup |
Troubleshooting
"MCP server not found" — Install uv: curl -LsSf https://astral.sh/uv/install.sh | sh
"AWS credentials not configured" — Run aws configure or aws sso login
"Access Denied" — Use ReadOnlyAccess policy or check permissions
License
MIT