The original title is: "MCPMark v2: InsForge MCP Uses 2.4x Fewer Tokens Than Supabase MCP on Claude Sonnet 4.6"

Short summary

InsForge reran its MCPMark benchmark on Claude Sonnet 4.6, comparing InsForge MCP against Supabase MCP across 21 Postgres tasks. InsForge achieved 28% higher Pass⁴ accuracy and used 2.4x fewer tokens, with the efficiency gap widening compared to Sonnet 4.5. The key finding: more capable models amplify the cost of incomplete backend context, making structured schema exposure increasingly important for agent efficiency.

• •InsForge MCP outperforms Supabase MCP by 28% Pass⁴ accuracy and 2.4x fewer tokens on Sonnet 4.6
• •The token-efficiency gap widened from 1.4x (Sonnet 4.5) to 2.4x (Sonnet 4.6), showing stronger models penalize missing context more
• •Structured backend context upfront reduces agent discovery queries, improving both accuracy and speed

Generated with AI, which can make mistakes.