Establish CPU, memory, and capacity baselines

[RickWinter]

Outcome

Maintainers and operators can quantify Azure MCP resource consumption and size local or remote deployments using reproducible evidence.

In scope

• Measure idle, startup, discovery, steady-state, and peak CPU and memory
• Track allocations, garbage collection, handles, and process growth where supported
• Measure resource behavior across server modes, tool counts, and concurrency levels
• Detect leaks and unbounded growth during sustained execution
• Produce capacity guidance for representative remote deployments

Out of scope

• Prescribing hosting SKU choices without workload assumptions
• Optimizing downstream Azure services
• Replacing hosting-platform monitoring

Acceptance criteria and budgets

• Controlled idle, steady-state, and peak resource baselines exist
• Steady-state CPU and memory may not increase by more than 10% from baseline
• Peak memory may not increase by more than 15%
• Sustained tests show no statistically significant unbounded memory, handle, or task growth
• Resource changes above 5% are reported with workload and attribution data
• Absolute resource envelopes are ratcheted after qualification

Dependencies and risks

• Controlled benchmark environments
• Resource-consumption telemetry work in Implement telemetry dashboard for resource consumption metrics #1567
• Remote-host operations Epic
• Performance Epic [EPIC] Performance #2412

Evidence for closure

• Resource benchmark suite
• Baseline trends
• Initial capacity and sizing guidance