Azure Pricing MCP Server

# Changelog All notable changes to the Azure Pricing MCP Server will be documented in this file. The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html). ## [3.3.0] - 2026-02-12 ### Added - **PTU Sizing + Cost Planner** (`azure_ptu_sizing` tool) - Estimate required Provisioned Throughput Units (PTUs) for Azure OpenAI / AI Foundry model deployments - Supports 19 models: gpt-5.2, gpt-5.1, gpt-5, gpt-5-mini, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, o3, o4-mini, gpt-4o, gpt-4o-mini, o3-mini, o1, Llama-3.3-70B-Instruct, DeepSeek-R1, DeepSeek-V3-0324, DeepSeek-R1-0528, and codex variants - Applies official rounding rules (minimum PTUs + scale increments per model and deployment type) - Supports Global, Data Zone, and Regional Provisioned deployment types - Accounts for output token multipliers (e.g., gpt-5: 1 output = 8 input tokens) - Supports cached token deduction (100% deducted from utilization per docs) - Optional live cost estimation via Azure Retail Prices API ($/PTU/hr, monthly projections) - Full calculation transparency: shows intermediate math, rounding rules, and data sources - Includes reservation guidance and benchmarking recommendations - **PTU Service** (`services/ptu.py`, `services/ptu_models.py`) - `PTUService` class with pure computation methods and async orchestrator - Versioned model data table sourced from official Microsoft PTU documentation - Case-insensitive model lookup with canonical name resolution ### Documentation - Added `azure_ptu_sizing` tool to TOOLS.md - Added PTU Sizing section to FEATURES.md ## [3.2.0] - 2026-02-10 ### Added - **Orphaned Resource Detection Tool** (contributed by [@iditbnaya](https://github.com/iditbnaya)) - `find_orphaned_resources` - Detect orphaned Azure resources and compute wasted costs - Scans for unattached managed disks, orphaned NICs, public IPs, NSGs, and empty App Service Plans - Integrates with Azure Cost Management API for historical cost lookup - Groups results by resource type with per-type summary tables - Configurable lookback period (default: 60 days) - Supports scanning all subscriptions or a single subscription - **Orphaned Resources Service** (`services/orphaned_resources.py`, `services/orphaned.py`) - `OrphanedResourceScanner` for async Resource Graph queries - Azure Cost Management integration for per-resource cost lookup - Uses existing aiohttp and azure-identity - no new dependencies ### Documentation - Added orphaned resource detection to TOOLS.md - Added detailed feature documentation in FEATURES.md - Added [@iditbnaya](https://github.com/iditbnaya) as contributor ## [3.1.0] - 2026-01-28 ### Added - **Spot VM Tools** (requires Azure authentication) - `spot_eviction_rates` - Query Spot VM eviction rates for SKUs across regions - `spot_price_history` - Get up to 90 days of Spot pricing history - `simulate_eviction` - Trigger eviction simulation on Spot VMs for resilience testing - **Azure Authentication Module** (`auth.py`) - `AzureCredentialManager` for Azure AD authentication - Non-interactive credential support (environment variables, managed identity, Azure CLI) - Graceful error handling with authentication help messages - Least-privilege permission guidance for each tool - **New Dependencies** - `azure-identity>=1.15.0` for Azure AD authentication (Spot VM tools) - **Spot Service** (`services/spot.py`) - Azure Resource Graph integration for eviction rates and price history - Azure Compute API integration for eviction simulation - Lazy initialization - auth only checked when Spot tools are called ### Configuration - `AZURE_RESOURCE_GRAPH_URL` - Resource Graph API endpoint - `AZURE_RESOURCE_GRAPH_API_VERSION` - API version for Resource Graph - `AZURE_COMPUTE_API_VERSION` - API version for Compute operations - `SPOT_CACHE_TTL` - Cache TTL for Spot data (1 hour default) - `SPOT_PERMISSIONS` - Least-privilege permission documentation ## [3.0.0] - 2026-01-26 ### ⚠️ Breaking Changes #### Entry Point Changed - **Console script entry point changed from `main` to `run`** - The `run()` function is now the synchronous entry point that wraps `asyncio.run(main())` - Existing console script configurations (`azure-pricing-mcp`) will continue to work - Code directly importing and calling `main()` still works (it's async) - This change improves the structure by clearly separating sync/async entry points #### `create_server()` Return Value - **`create_server()` now returns a tuple `(Server, AzurePricingServer)` by default** - This change exposes the pricing server for testing and advanced use cases - Use `create_server(return_pricing_server=False)` for the previous behavior (returns only `Server`) - The `AzurePricingServer` instance is needed for lifecycle management #### Session Lifecycle Management - **HTTP session is now managed at the server level, not per-tool-call** - Previously: Each tool call created and destroyed a new HTTP session (inefficient) - Now: A single HTTP session is created at server startup and reused for all tool calls - This significantly improves performance and reduces overhead - When using `AzurePricingServer` directly, you must manage its lifecycle: ```python # Option 1: Context manager (recommended) async with AzurePricingServer() as pricing_server: result = await pricing_server.tool_handlers.handle_price_search(...) # Option 2: Manual lifecycle management pricing_server = AzurePricingServer() await pricing_server.initialize() try: result = await pricing_server.tool_handlers.handle_price_search(...) finally: await pricing_server.shutdown() ``` ### Added - **Modular Services Architecture** - `client.py` - HTTP client for Azure Pricing API - `services/` - Business logic (PricingService, SKUService, RetirementService) - `handlers.py` - MCP tool routing - `formatters.py` - Response formatting - `models.py` - Data structures - `tools.py` - Tool definitions - `config.py` - Configuration constants - **New `AzurePricingServer` Methods** - `initialize()` - Explicitly start the HTTP session - `shutdown()` - Explicitly close the HTTP session - `is_active` property - Check if session is active - **Improved Documentation** - Comprehensive docstrings for all public APIs - Breaking change documentation in module docstring ### Changed - Restructured codebase from monolithic to modular architecture - Updated all tests to use service-based architecture with proper dependency injection - Improved error handling with session state checks ### Removed - Obsolete documentation files: - `DOCUMENTATION_UPDATES.md` - `MIGRATION_GUIDE.md` - `QUICK_START.md` (replaced by README quick start section) - `USAGE_EXAMPLES.md` (replaced by README examples) ### Migration Guide #### For Console Script Users No changes required. The `azure-pricing-mcp` command continues to work. #### For Library Users 1. **If you call `create_server()`:** ```python # Old (v2.x) server = create_server() # New (v3.0) - if you don't need pricing_server server = create_server(return_pricing_server=False) # New (v3.0) - if you need pricing_server for testing server, pricing_server = create_server() ``` 2. **If you use `AzurePricingServer` directly:** ```python # You MUST initialize the session before tool calls async with AzurePricingServer() as pricing_server: # All tool calls within this block share the same HTTP session result = await pricing_server.tool_handlers.handle_price_search(...) ``` ## [2.3.0] - Previous Release See git history for changes in previous versions.