Scalpel – give AI coding agents get_symbol() instead of a whole file (github.com)

A new tool called Scalpel has been introduced, allowing AI coding agents to retrieve specific code definitions and their usages through a streamlined command, `get_symbol(name)`, instead of processing entire files. This innovation results in substantial efficiency gains, with the tool demonstrating a 90% reduction in token usage when answering queries about function definitions. Scalpel operates via a Minimal Computation Protocol (MCP) server, significantly enhancing the performance of coding agents by providing precise information without unnecessary data overhead. This development is particularly noteworthy for the AI/ML community as it suggests a major advancement in code retrieval techniques. Scalpel's performance metrics show it outperforms established competitors like grep and jcodemunch in both definition accuracy and token efficiency. For instance, while grep requires over 6,700 tokens for definition lookups, Scalpel achieves the same accuracy for just 1,732 tokens. Additionally, its effectiveness was validated through tests against 57 real historical bug fixes, indicating that while it does not enhance the overall speed of task completion, it offers clear benefits in environments where token cost is a significant factor and for tasks requiring frequent reference lookups. Scalpel underscores a direction toward more resource-efficient AI tools in software development workflows.