Expanding Managed Agents in Gemini API: background tasks, remote MCP and more

Source: Google Date: 2026-07-07 URL: https://blog.google/innovation-and-ai/technology/developers-tools/expanding-managed-agents-gemini-api/

Summary

Google (Philipp Schmid, Mariano Cocirio at DeepMind) expanded Managed Agents in the Gemini API with background/async execution (no held-open HTTP connections), direct remote MCP server integration, custom function calling alongside sandbox tools, and mid-session credential refresh — all via the existing Gemini Interactions API and @google/genai SDK.

Implications

• code-mode-over-MCP — Google wiring remote MCP directly into its managed-agent runtime (rather than requiring custom middleware) is a data point on the opposite side of the code-mode debate: first-class MCP support persisting as a hosted-agent primitive even as some harnesses move orchestration into code.
• agent-layer convergence — background execution, external tool credentials, and remote tool servers are the same primitive set Anthropic and OpenAI are independently converging on for managed/hosted agents.