Arcade frames agent actions around authorization and governance

Agent products increasingly need to turn intent into real action across user-owned SaaS accounts, but raw API keys and unconstrained tool calls are a weak production boundary. Arcade's documentation describes the product as an enterprise-ready actions runtime for AI agents, with guides and references for agent authorization, agent-optimized tools, MCP servers, sample agents, and agent lifecycle governance. The docs homepage also points to a public `arcade-mcp` repository, dashboard, status surface, examples, and release channels, which makes the runtime visible as infrastructure rather than a hidden integration library. For agent builders, the practical signal is to evaluate action layers by whether they separate user authorization from model prompts, expose a governed tool catalog, support MCP-compatible integration, provide sample-agent implementation paths, surface status and lifecycle controls, and let agents act on behalf of users without scattering credentials through custom glue code.