One CLI built from Google's discovery docs vs one MCP tool per API: what works for my agents
retained text
For my agents, I use one CLI instead of one MCP tool per API, and it works better for me. Google Workspace has dozens of APIs, and a tool per API means dozens of tool definitions for the agent to carry. I use Google's open-source gws CLI (googleworkspace/cli), which builds its command tree at runtime from Google's discovery docs. The agent looks up a method's schema first, writes run with --dry-run to preview the request, and field masks keep responses small. One lesson: if your OAuth app is in Testing mode, refresh tokens expire every 7 days, so a scheduled agent stops authenticating after a week. For agent tooling, do you prefer a CLI or MCP tools, and why?
document fetched 2026-10-20 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1x9pzmt/one_cli_built_from_googles_discovery_docs_vs_one