Dev Tools · 1h ago
InsForge MCP beats Supabase on Sonnet 4.6 with 2.4x fewer tokens
InsForge MCP achieved 42.86% Pass⁴ accuracy on MCPMark v2 benchmarks using Claude Sonnet 4.6, compared to Supabase MCP's 33.33%. InsForge used 7.3M tokens per run versus Supabase's 17.9M, a 2.4x efficiency gap that widened from 1.4x on Sonnet 4.5. The benchmark tested 21 database tasks with strict repeated execution scoring.
Meridian48 take
The token efficiency gap widening on newer models suggests that surfacing backend context upfront is becoming more critical as models reason more extensively when information is incomplete.
mcp-benchmarkdatabase-tools