I Built an MCP Server That Gives AI Agents Real-Time AWS Pricing Data

#aws #finops #ai #mcp

How I solved the 'how much will this cost?' problem for AI-powered infrastructure tools

The Problem

You're building an AI agent that provisions cloud infrastructure. User says: "Spin up a t3.medium in the cheapest region."

Your agent has no idea what that costs. Or which region is cheapest. Or whether reserved instances would save money.

I kept hitting this wall while building AI-powered DevOps tools. The agent could do things, but couldn't answer the most basic question: "How much will this cost?"

The Solution: AWS Pricing MCP

I built an MCP-compatible API that returns real-time AWS pricing data. It's live on RapidAPI now.

What it does:

Query pricing for 20 AWS services (EC2, RDS, Lambda, S3, DynamoDB, ECS, EKS, etc.)
Compare prices across all 18 AWS regions
Get on-demand vs reserved instance breakdowns
Estimate monthly costs for multi-instance workloads

How It Works

The API follows MCP (Model Context Protocol) conventions, so any MCP-compatible AI client can use it directly.

Basic Query

curl -X GET "https://aws-pricing-mcp.p.rapidapi.com/ec2/pricing?instance_type=t3.medium&region=us-east-1" \
  -H "X-RapidAPI-Key: YOUR_KEY"

Response

{
  "instance_type": "t3.medium",
  "region": "us-east-1",
  "prices": {
    "on_demand_hourly": 0.0416,
    "on_demand_monthly": 30.37,
    "reserved_1yr_no_upfront_monthly": 24.82,
    "reserved_1yr_all_upfront_monthly": 21.90,
    "reserved_3yr_all_upfront_monthly": 14.60
  },
  "vcpu": 2,
  "memory_gb": 4,
  "potential_savings": {
    "reserved_1yr_percent": 27.9,
    "reserved_3yr_percent": 51.9
  }
}

Region Comparison

This is where it gets useful. Find the cheapest region for any instance type:

curl -X GET "https://aws-pricing-mcp.p.rapidapi.com/ec2/compare-regions?instance_type=m5.xlarge"

{
  "instance_type": "m5.xlarge",
  "cheapest_region": "us-east-2",
  "cheapest_price_monthly": 140.16,
  "most_expensive_region": "ap-northeast-1",
  "most_expensive_price_monthly": 185.04,
  "potential_savings_percent": 24.2,
  "all_regions": [
    {"region": "us-east-2", "monthly": 140.16},
    {"region": "us-east-1", "monthly": 142.08},
    ...
  ]
}

Multi-Service Cost Estimation

Estimate costs for a complete workload:

curl -X POST "https://aws-pricing-mcp.p.rapidapi.com/estimate" \
  -H "Content-Type: application/json" \
  -d '{
    "region": "us-west-2",
    "resources": [
      {"service": "ec2", "instance_type": "t3.medium", "count": 3},
      {"service": "rds", "instance_type": "db.t3.medium", "count": 1},
      {"service": "s3", "storage_gb": 100}
    ]
  }'

Integration with AI Agents

If you're building with Claude, GPT, or any MCP-compatible system, the API responses are structured for easy parsing. No wrestling with AWS's pricing API directly.

Example prompt your users might give:

"What's the cheapest way to run a 3-node web tier with a managed database in AWS?"

Your agent can now actually answer that with real numbers.

Services Supported

Service	Pricing Data
EC2	On-demand, Reserved (1yr/3yr), Spot
RDS	All engines, Multi-AZ
Lambda	Request + duration
S3	Storage classes, transfer
DynamoDB	On-demand + provisioned
ECS/EKS	Fargate + EC2 launch types
ElastiCache	Redis + Memcached
OpenSearch	Instance + storage
...	20 services total

Pricing

Tier	Requests/Month	Price
Free	100	$0
Pro	5,000	$49
Ultra	25,000	$149
Mega	100,000	$299

The free tier is enough to prototype and test.

Why I Built This

Two reasons:

AWS's pricing API is painful. The official API requires you to construct complex filter queries and returns data in a format optimized for AWS, not developers.
AI agents need structured data. When you're building tools that AI can use, you need clean JSON with predictable schemas. Not XML. Not HTML scraping. Not "check the console."