Benchmarks

Performance that speaks for itself.

Comprehensive benchmarks (July 2026) demonstrate WhiteMagic delivers 3-10x faster response times than typical MCP implementations, with 100% success rates and minimal resource usage.

Median latency

29-33ms

3-10x faster than typical (100-300ms)

P95 latency

36-55ms

95th percentile response time

Success rate

100%

Zero failures across all tests

Memory per call

0-0.18MB

Minimal resource footprint

Tool-by-tool benchmarks

Tool	Median (ms)	P95 (ms)	P99 (ms)
list_ganas	29.30	38.23	56.15
vitality	29.64	36.45	39.97
discover	31.71	41.99	62.22
gnosis	30.35	38.64	40.40
karma_report	31.14	35.18	50.89
capability.matrix	32.90	39.63	82.02
health_report	31.98	40.49	58.17
state.summary	32.44	41.00	55.80

How WhiteMagic compares

System	Median (ms)	P95 (ms)	Notes
WhiteMagicBest	29-33	36-55	Lean, optimized
Anthropic Reference	100-300	200-500	Standard implementation
Popular MCP Servers	50-150	100-300	Filesystem, git, etc.
Complex Frameworks	200-1000	500-2000	LangChain, CrewAI
Database Tools	100-500	300-1000	SQL, vector DBs

Security systems working as designed

Concurrent load testing revealed WhiteMagic's security systems (circuit breaker and rate limiter) successfully protect against abuse while maintaining fast response times for legitimate users.

Sequential throughput: 29.38 requests/second
Concurrent (5 workers): 5.59 requests/second
Concurrent (10-20 workers): ~4 requests/second
Security: Circuit breaker and rate limiter blocked rapid-fire calls automatically

Why WhiteMagic is so fast

Lean architecture: No FastMCP overhead, direct SDK usage
Lazy imports: Heavy modules deferred until first use
Efficient dispatch: Optimized PRAT routing table
Local execution: Most tools run locally without external API calls
Minimal memory: 0-0.18MB per call (vs. 10-100MB typical)

Benchmark methodology

Benchmarks conducted on July 29, 2026 using 50 iterations per tool with 5 warmup calls. Tests run on production hardware with all 860 tools loaded. Full benchmark suite available at core/benchmark_results/benchmark_report.md.