Two of four code-Intel MCPs scored below grep in a benchmark I built (github.com)

In a recent benchmark comparing five AI coding agents for navigating unfamiliar codebases, the tool Sense emerged as the clear winner, outperforming competitors in all six tested scenarios. Sense notably excelled in fairness, citation grounding, and efficiency while also being the fastest and least expensive option. The comparisons indicate that Sense, which utilizes a symbol graph and embeddings over the MCP (Multi-Context Protocol), effectively combines structural and semantic search capabilities, making it highly efficient for developers needing swift insight into complex codebases. This evaluation is significant for the AI/ML community as it highlights the operational benefits of AI tools like Sense in software development, particularly in enhancing code comprehension and modification processes. The scoring metrics specified a comprehensive approach, weighing factors like keyword coverage and answer quality, making it easier to understand the effectiveness of these tools. The results could influence the adoption of coding AIs and drive improvements in their underlying technologies, as tools that leverage efficient indexing and provide high-quality citations are crucial for reducing technical debt and improving developer productivity.