Go 1.1x faster.
1.1x faster.
Zero dependencies.
ZeroMCP vs mcp-go (mark3labs) — HTTP benchmarks.
Throughput
4,740 req/s ZeroMCP
vs. 4,199 req/s mcp-go
Latency (p50)
0.18 ms ZeroMCP
vs. 0.17 ms mcp-go
Memory
30 MB ZeroMCP
vs. 14 MB mcp-go
HTTP Performance — Head to Head
Same hello tool. Same methodology. 5-minute sustained load in Docker. Chi router for ZeroMCP, built-in HTTP for mcp-go.
req/s p50 p99 Memory CPU
ZeroMCP (Chi) 4,740 0.18ms 0.45ms 30 MB 0.00%
mcp-go (mark3labs) 4,199 0.17ms 0.67ms 14 MB 0.00%
The tradeoff
mcp-go (mark3labs) is a community Go library with its own HTTP transport baked in — lean at 14 MB. ZeroMCP is a zero-dependency handler you mount on Chi, Gin, or net/http, which is why it pushes 4,740 req/s to mcp-go's 4,199 but uses more memory (30 MB vs 14 MB) from the router overhead. Neither is wrong. mcp-go is simpler to keep lightweight. ZeroMCP is simpler to embed in an existing Go service.