Performance benchmarks for MCP server

[elijahr]

Problem

The MCP server is the backbone of spellbook's runtime functionality. There are no benchmarks measuring tool response times, startup latency, or throughput under load.

Proposed Benchmarks

Startup Time

• Time from spellbook start to first tool availability
• Measure with different numbers of skills loaded

Tool Response Time

• Simple tools (e.g., spellbook_health_check): target <50ms
• Complex tools (e.g., skill_instructions_get with large skills): target <200ms
• Tools with file I/O (e.g., workflow_state_save): target <500ms

Throughput

• Concurrent tool calls
• Memory usage under sustained operation

Regression Detection

• Store benchmark results in CI
• Alert on >20% regression

Implementation

Use pytest-benchmark or a custom benchmark harness. Results can be visualized in the docs site (similar to how Ruff and uv display benchmark charts).