Show HN: LTP – Lazy Tool Protocol. Up to 93% token reduction for AI agents (github.com)

A new system called Lazy Tool Protocol (LTP) has been introduced, enabling AI agents to significantly reduce token consumption by up to 93% when accessing tools. Unlike traditional methods that consume large amounts of tokens for tool definitions, LTP allows on-demand tool discovery using simple command-line interface (CLI) commands. This means that where a traditional Model Context Protocol (MCP) call for 300 tools would typically use around 60,000 tokens, LTP utilizes zero tokens for tool discovery, drastically cutting costs and making the process more efficient. This advancement is crucial for the AI/ML community, as it enhances the efficiency of AI tasks by lowering the cost and complexity associated with tool usage. LTP introduces reusable workflows, termed "Crafts," which integrate instructions with executable scripts, expanding functionality beyond mere static commands. The protocol exhibits a unique Lazy Loading Pattern that optimally reduces token consumption, especially with repetitive calls, potentially converging to a reduction of 85-90%. By facilitating a seamless connection to any MCP server and promoting valuable resource management, LTP promises to boost productivity and foster innovation in AI development processes.