Gemini CLI Extensions for Figma (aicloudlab.substack.com)

Google has rolled out Gemini CLI extensions, a new framework that lets developers plug external tools into the Gemini CLI—bringing the Gemini model directly into terminal workflows. The Figma extension demonstrates the concept by linking Figma to Gemini’s Reason-and-Act (ReAct) agent so designers’ layouts can be converted into working code without leaving the terminal. Extensions are modular “power-ups” that include a playbook (metadata describing MCP server settings, available commands and permissions) so Gemini immediately understands how to use the connected tool. At the time of writing the walkthrough uses Gemini CLI v0.8.2 and installs extensions under .gemini/extensions/<mcpServername>; Figma auth is done via /mcp auth figma with OAuth tokens persisted in ~/.gemini/mcp-oauth-tokens.json. For the AI/ML community this formalizes a practical pattern for making LLMs first-class actors in developer environments: an LLM agent that can call APIs, manipulate design artifacts, run local tools, and interact with Model Context Protocol (MCP) servers. That reduces context-switching, accelerates design-to-code iteration, and enables more reproducible agent behavior because each extension’s playbook codifies capabilities and permissions. Expect broader impact as more extensions connect databases, cloud services, monitoring, and CI systems—turning LLMs into extendable automation hubs while raising the usual considerations around auth, permissions, and secure token storage.