GAX – A tool execution protocol to fix MCP token bloat and CLI auth (github.com)

GAX, or Governed Agent eXecution, has been introduced as a new open protocol designed to streamline how AI agents interact with external tools. Unlike existing methods like the Model Context Protocol (MCP), which can impose heavy token usage and operational complexity, GAX combines the efficiency of command-line interfaces (CLI) with advanced governance features. By utilizing a lightweight approach that includes a sidecar for managing OAuth, policy, and audits, GAX allows agents to invoke commands with minimal upfront token costs while maintaining robust security standards. This innovation is significant for the AI/ML community as it addresses critical pain points in automated tool execution, particularly in environments requiring high efficiency, such as GitHub, Kubernetes, and various SaaS APIs. The architecture of GAX facilitates lazy discovery of commands and capabilities on demand, which can significantly reduce token overhead—resulting in performance gains of up to 32 times compared to naive MCP setups. By providing a structured output envelop that includes pertinent metadata, GAX enhances the audibility of operations and supports seamless integration across different tools, paving the way for more sophisticated AI-driven automation.