ZeroMCP v0.1.1 — ~27% faster with schema caching. See what changed →
v0.1.1

~27% faster. One line changed.

April 7, 2026

Schema caching eliminates per-request toJsonSchema() recomputation.
All 10 languages. One optimization. Measurable everywhere.

What changed

Schema caching

Tool JSON Schema is now computed once at registration and cached. Previously recomputed on every tools/list and tools/call request.

Go: HandleRequestBytes

New byte-level request handler eliminates double JSON serialization in Go. Requests go straight from []byte to response without intermediate marshal/unmarshal.

Hono memory leak fixed

The v0.1.0 Hono memory growth (122 MB to 739 MB over 5 min) was caused by per-request schema allocation. Schema caching eliminated it entirely.


Performance impact

Same methodology as v0.1.0: Docker containers, HTTP transport, 30-second sustained load. Official SDK numbers unchanged.

Language Framework v0.1.1 v0.1.0 Improvement vs Best Third-Party
Rust Actix 5,879 req/s 5,111 +15%
Node.js bare http 5,497 req/s 4,539 +21% 2.4x vs mcp-framework
C# ASP.NET 5,260 req/s 4,421 +19%
Ruby Rack+Puma 4,762 req/s 3,217 +48% 1.3x vs fast-mcp
Go Chi 4,740 req/s 4,024 +18% 1.1x vs mcp-go
Java Javalin 4,512 req/s 3,791 +19% 1.9x vs spring-ai-mcp
Python Starlette 3,528 req/s 2,623 +35% 2.0x vs mrexodia
Kotlin Ktor 3,163 req/s 2,848 +11%
Swift Vapor 2,244 req/s 1,730 +30%
PHP Slim 1,726 req/s 1,561 +11%

Ruby saw the largest gain (+48%) — its per-request schema computation was the most expensive relative to request handling.


Node.js framework update

Schema caching fixed the Hono memory leak and improved all frameworks. 5-minute sustained load.

Framework v0.1.1 req/s Memory CPU
bare http 5,065 24-28 MB 0.03%
Fastify 4,877 85-86 MB 0.04%
Hono 4,385 75-76 MB 0.04%
Express 4,017 74-78 MB 0.03%

Hono: stable at 75-76 MB for the full 5 minutes. No more memory growth.

Upgrade

Drop-in replacement. No API changes. No config changes. Just update the version.