Measure command dispatch and transport overhead

[RickWinter]

Outcome

Azure MCP can quantify and control the latency it adds before and after downstream Azure operations across stdio and remote HTTP transports.

In scope

• Benchmark command binding, validation, routing, serialization, and response handling
• Measure stdio and HTTP transport overhead independently
• Use deterministic no-op or local test services to exclude Azure and network latency
• Compare namespace, consolidated, and supported routing modes where applicable
• Report overhead distributions by execution layer

Out of scope

• Optimizing downstream Azure SDK or service latency
• Measuring tool-selection accuracy
• End-to-end customer network performance

Acceptance criteria and budgets

• Server-only dispatch and response overhead have controlled p50, p95, and p99 baselines
• p50 and p95 may not regress by more than 10% from the approved baseline
• p99 may not regress by more than 20%
• Transport-specific overhead changes above 5% are reported
• Results identify the responsible execution layer for a material regression
• Absolute per-transport overhead targets are ratcheted after qualification

Dependencies and risks

• Deterministic benchmark commands and local service doubles
• Stable transport and routing implementations
• Performance Epic [EPIC] Performance #2412

Evidence for closure

• Layered benchmark suite
• Approved baselines
• Transport comparison and absolute targets