Azure Cost Calculator - AI Agent Plugin

Real-time Azure cost estimation using the public Azure Retail Prices API. Works with any agent in the skills.sh ecosystem. All prices come from live API lookups. No Azure subscription required.

Install

Install as a Plugin (recommended for Copilot CLI and Claude Code)

Run inside a Copilot CLI or Claude Code session.

/plugin marketplace add ahmadabdalla/plugins
/plugin install azure-cost-calculator@ahmadabdalla-plugins

Plugin install pulls from versioned releases with changelog tracking and update control. For the full plugin install/uninstall/update guide, see Plugin Management.

Install as a Skill (works with any agent)

npx (works with any agent):

npx skills add ahmadabdalla/azure-cost-calculator-skill

Don't have npx? Install Node.js (which includes npm and npx), or run npm install -g skills first then use skills add ahmadabdalla/azure-cost-calculator-skill.

The npx method pulls the latest from the main branch directly; it always gets the current stable content but without version pinning or rollback.

Usage

Ask about Azure costs in natural language. The skill activates automatically.

How much does a D4s v5 VM cost per month in East US?
Compare App Service pricing tiers for a production web app
Estimate a Standard_B2s VM with a P30 managed disk in Australia East in AUD
What's the cost of a General Purpose SQL Database with 4 vCores in West Europe in EUR?
How much would Azure Cosmos DB with 1000 RU/s and 100 GB storage cost?

Planning a larger architecture? Start with:

I'd like to perform a cost analysis on an Azure architecture. What do I need to consider to get consistent results?

The agent will walk you through the key parameters that affect pricing accuracy. For the full guide on writing prompts that produce deterministic estimates, see the Usage Guide.

How It Works

The skill uses service reference files as an index. Each file contains exact API filter values as declarative Key: Value parameters, cost formulas, and traps. The agent reads the matching file, translates the parameters to the detected runtime (Bash or PowerShell), runs the pricing script against the live API, and presents a structured estimate.

The skill optimises for two goals:

• Determinism - target ≤ 5% cost variance. Same query → same API call → same price. All values from the live API, nothing hardcoded. LLMs are non-deterministic by nature, so this skill is designed to constrain them where possible: pre-verified filters, explicit formulas, and scripted API calls reduce the surface area where the model can drift.
• Token efficiency - target ≤ 5% token usage variance. Only SKILL.md and shared.md load on every query. Service files load on demand. Batch mode (3+ services) reads only the first 45 lines per file.

Other design goals:

• Multi-currency, all regions - supports USD, AUD, EUR, GBP, JPY, CAD, INR, etc. Works with any Azure region.

Note: Targets measured via A/B testing with clean-context sessions against complex Azure architectures. Tested with Claude Opus 4.6 and Gemini Pro 3. Results with other models may vary.

References load on demand, keeping token usage low even for 10+ service estimates.

Supported Services

210+ Azure services are mapped across 18 categories (Compute, Databases, Networking, Storage, Security, Monitoring, Integration, AI + ML, and more). 110+ services have full reference files with documented query patterns. For services without a reference file, the skill includes an exploration script that searches the live API to find the right filters automatically.

Found a Gap? Open an Issue

If you query a service and the skill falls back to discovery mode, that's a signal we're missing a reference file. Please open an issue with the service name rather than accepting the best-effort result. Even if the estimate looked correct this time, the next user (or the next API change) may not get the same result. Issues help us prioritise which reference files to add next.

Prerequisites

• Bash with curl and jq (macOS/Linux, preferred), or PowerShell 7+ (pwsh); install on Windows/macOS/Linux. Windows ships with PowerShell 5.1 (powershell.exe) which is not the same as pwsh; you must install PowerShell 7 separately.
• Internet access to https://prices.azure.com
• No Azure subscription or authentication required

Contributing

Each service reference file you add improves accuracy for everyone. See CONTRIBUTING.md for the full guide, including a ready-to-use prompt that walks your AI through generating a complete reference file.

FAQ

Why is this free?

1. The API is free. The Azure Retail Prices API is public and requires no authentication or Azure subscription.
2. We can't guarantee determinism. LLMs are non-deterministic by nature. While this skill constrains the model for consistency, we can't commit to identical results for identical prompts, and we don't think it's right to charge for something we can't make fully deterministic.
3. Community-driven by design. Some things are meant to be open source because they rely on community support to grow. This project is one of them, and every service reference file contributed improves accuracy for everyone.

Why contribute service references?

Without a reference file the agent still works, but it has to discover API filter values on the fly, using more tokens and risking inconsistent results. A reference file provides pre-verified query patterns, documented traps, and correct cost formulas, so every user gets consistent results.

	With reference file	Without (discovery mode)
API query	Pre-verified filters	Agent discovers filters from the live API
Known gotchas	Documented - the agent avoids common pricing quirks automatically	Agent still works, but may not catch edge cases like $0.00 rounding or RI math
Multi-part costs	Each component (compute, storage, IP, etc.) has its own query	Agent queries the main component; secondary costs may need a follow-up
Cost formula	Correct multipliers, free-tier deductions, tiered pricing	Uses the API's unit of measure - usually right, occasionally off for unusual meters
Speed	Fast - fewer tokens	Slower - requires a discovery step first
Accuracy	High - patterns tested against the live API	Depends on model quality - may vary without pre-verified patterns

Should I install as a global or project skill?

Global installs the skill once and makes it available across all your projects without duplicating files. Project-scoped installs it inside a single project. Use global if you estimate Azure costs regularly; use project if you only need it for a specific repo.

License

This project is licensed under the MIT License.

Support

If you find this skill useful, consider buying me a coffee: