aws-calculator-mcp 1.2.2

Turn a list of AWS services into an official, shareable AWS Pricing Calculator link with real costs baked in. MCP server + REST API + CLI.

AWS Pricing Calculator — Estimate Generator

Describe your AWS setup in plain English and get back an official, shareable AWS Pricing Calculator link — a real https://calculator.aws/#/estimate?id=... with real costs already computed and baked in. No AWS account, no credentials, no manual clicking, no JSON to write.

$ aws-calc --prompt "2 t3.large EC2 with 50GB, RDS MySQL db.m5.large 100GB, 500GB S3, an ALB"

✅ Monthly cost: $312.40 USD
🔗 https://calculator.aws/#/estimate?id=8422dc22a2849dbf798ab405385a7e5f32fcb055

Use it three ways — all from one install:

• CLI (interactive or one-liner) — for anyone, scripts, CI, .exe
• MCP server — inside Claude, Cursor, VS Code, any MCP client
• REST API (optional) — for ChatGPT actions, automation, other languages

Self-contained: the module handles cost-baking itself — the browser engine it needs is auto-downloaded on first run. You install one pip package; nothing else.

---

Contents

1. Install
2. Quick start — plain English
3. Interactive mode
4. All CLI commands
5. Use it in Claude / Cursor / IDEs (MCP)
6. Use it from ChatGPT / automation (REST)
7. Supported services & config
8. Commitment pricing (1yr / 3yr) & auto-scaling
9. How it works
10. Limitations

---

1. Install

pipx install aws-calculator-mcp        # recommended (isolated)
# or
pip install aws-calculator-mcp

That's it. On the first estimate it downloads the Chromium engine once (~150 MB) to compute costs; every run after is instant. (Debian/Ubuntu "externally managed" error with pip? use pipx, or add --break-system-packages.)

2. Quick start — plain English

aws-calc --prompt "3 t3.large EC2 with 50GB each, an RDS MySQL db.m5.large 100GB \
multi-az, a 500GB S3 bucket, CloudFront 1TB, 10 lambdas with 2M requests, an ALB \
and an NLB in Mumbai"

It shows exactly what it understood, then returns the link with costs baked in:

   Understood from your prompt:
     • ec2 (instances=3, instance_type=t3.large, storage_gb=50)
     • rds mysql (instance_type=db.m5.large, storage_gb=100, deployment=multi-az)
     • s3 (storage_gb=500)
     ...
   Monthly cost: $782.00 USD
   🔗 https://calculator.aws/#/estimate?id=...

Add --group to organise the estimate into categories (Compute, Database, …). Misspellings are tolerated (lamda, buckit, dynmodb, cloud front all work).

3. Interactive mode

Don't want to remember flags? Just run:

aws-calc -i        # or simply `aws-calc` with no arguments

It walks you through it:

Estimate name [My Estimate]: Prod Stack
AWS region (e.g. us-east-1, ap-south-1) [us-east-1]: ap-south-1
Group services by category? (y/n) [y]: y
Describe your infrastructure: 2 m5.large EC2, RDS MySQL db.m5.large 100GB, 500GB S3, an ALB
→ 🔗 https://calculator.aws/#/estimate?id=...

4. All CLI commands

# plain-English prompt
aws-calc --prompt "2 m5.large ec2, rds mysql db.m5.large 100gb, 1tb s3"

# group into categories, pin a region
aws-calc --group --region ap-south-1 --prompt "ec2, rds, s3, alb, waf"

# interactive walkthrough
aws-calc --interactive

# one explicit service (full control)
aws-calc --service EC2 --region us-east-1 \
         --config '{"instances":2,"instance_type":"t3.large","storage_gb":50}'

# from a JSON file (full control over groups/services)
aws-calc --file estimate.json

# pipe a sentence in, or JSON in — JSON out for scripts/CI
echo "3 t3.medium ec2 and 1tb s3" | aws-calc --json
cat estimate.json | aws-calc --json

# fast draft link (skip cost-baking)
aws-calc --prompt "..." --no-costs

# name it
aws-calc --name "Client Proposal" --prompt "..."

Flags: --prompt/-p, --interactive/-i, --group, --region/-r, --service/-s, --config/-c, --file/-f, --name/-n, --no-costs, --json.

5. Use it in Claude / Cursor / IDEs (MCP)

After pipx install aws-calculator-mcp, add to your MCP config (claude_desktop_config.json, Cursor, VS Code Continue, Windsurf, …):

{
  "mcpServers": {
    "aws-calculator": {
      "command": "aws-calc-mcp"
    }
  }
}

Zero-install variant (via uv): "command": "uvx", "args": ["--from","aws-calculator-mcp","aws-calc-mcp"].

Then just chat: "Create an AWS estimate: 3 t3.large web servers, an RDS MySQL db.m5.large with 100 GB, a 500 GB S3 bucket, grouped by category." The model calls the create_estimate tool (which accepts structured services or a prompt, plus group) and returns the link.

6. Use it from ChatGPT / automation (REST)

Optional — run the REST API yourself (only if you want an HTTP endpoint):

pip install "aws-calculator-mcp[api]"
aws-calc-api                 # serves on :8080  (honors $PORT)

curl -X POST http://localhost:8080/v1/estimate -H "Content-Type: application/json" \
  -d '{"prompt": "2 m5.large EC2, RDS MySQL db.m5.large 100GB, 500GB S3", "group": true}'

Endpoints: GET /health, GET /v1/services, GET /v1/regions, POST /v1/estimate. Interactive docs at /docs; OpenAPI at /openapi.json (import that into a ChatGPT Custom GPT Action to use it from ChatGPT). Point other installs at it with export AWS_CALC_API_URL=http://your-host:8080 so they skip local baking.

7. Supported services & config

~50 services, tested against AWS's own pricing engine:

Category	Services
Compute	EC2, Lambda, Fargate, EKS, Lightsail
Storage	S3, EBS, EFS, ECR
Database	RDS (MySQL, PostgreSQL, Oracle, SQL Server, MariaDB), Aurora (MySQL/PostgreSQL), DynamoDB, Redshift, OpenSearch, ElastiCache (Redis/Valkey/Memcached)
Network	CloudFront, Route 53, API Gateway, ELB/ALB/NLB, VPC, Network Firewall, Site-to-Site VPN, NAT Gateway, Transit Gateway, PrivateLink
Security	WAF, GuardDuty, KMS, Cognito, Inspector, Security Hub
Mgmt	CloudWatch, CloudTrail, Config
Messaging	SQS, SNS, SES, Kinesis
AI / DR	Bedrock, Elastic Disaster Recovery (EDR/DRS)
Dev	CodeBuild

When writing JSON (--file or the MCP/REST services/groups), each service is {"service","region","description","config"}. Common config keys:

EC2:        instances, instance_type, os, storage_gb, pricing, term, upfront, hours_per_day
Lambda:     requests, duration_ms, memory_mb
S3:         storage_gb, get_requests, put_requests, data_returned_gb
RDS *:      instance_type, storage_gb, deployment (single-az|multi-az)
Aurora:     engine (mysql|postgresql), nodes, instance_type
DynamoDB:   mode (provisioned|on-demand), read_capacity, write_capacity, storage_gb
ElastiCache:engine, nodes, node_type
CloudFront: data_transfer_gb, https_requests
API Gateway:http_requests_million
ALB/NLB:    load_balancers, data_processed_gb

Full list: GET /v1/services, or ask an MCP client to "list services".

8. Commitment pricing & auto-scaling

{"service":"EC2","config":{
  "instance_type":"m5.large", "instances":2,
  "pricing":"compute-savings",  // or instance-savings | reserved | on-demand | spot
  "term":"3yr",                 // 1yr | 3yr
  "upfront":"all"               // none | partial | all
}}

Verified: m5.large ×2 — On-Demand $140 → Savings 1yr $103 → 3yr $71. Model part-time / auto-scaling fleets with "hours_per_day": 5 or "utilization": 30. RDS/Aurora/Redshift/ElastiCache accept "pricing":"reserved".

9. How it works

The AWS calculator computes prices client-side in the browser and only stores the numbers in a saved estimate — there's no public compute endpoint. So this tool: builds the exact payload AWS expects (services.py) → POSTs it to AWS's saveAs API → opens the draft in a headless browser, clicks Update estimate so AWS's own engine computes costs, then Share to re-save with them baked in (compute.py). Because it uses AWS's engine, numbers always match calculator.aws.

parser.py    plain English → services            core.py     build → save → bake
services.py  service + config → AWS payload       compute.py  headless-browser baking
server.py    MCP (stdio)   api_server.py REST     cli.py      CLI + interactive

10. Limitations

• A few services may show $0 until you set a value on the page after opening: AWS Backup (nested form the save API can't reach), Transfer Family, CodePipeline (near-free), EC2 standard Reserved (use Savings Plans), DynamoDB on-demand (provisioned works).
• Bedrock uses the default Amazon model (token-rate based).
• The natural-language parser is heuristic — name services explicitly for best results (an MCP client like Claude parses free-form prose more flexibly).
• Cost-baking runs a headless browser locally (auto-installed). On a server, that's ~1–2 GB RAM during baking. If a host can't run it, you still get a working draft link (costs appear after one "Update estimate" click).

---

Issues & PRs: github.com/vireshsolanki/aws-calculator-mcp · MIT licensed · not affiliated with AWS (uses the public calculator.aws endpoints).