Google Deepmind adds background execution and MCP support to Gemini API managed agents

What changed

Google Deepmind expanded the Gemini API Managed Agents with four practical new features. Agents can now run tasks asynchronously in the background, allowing workflows to continue without waiting for immediate responses. They also support direct connections to remote Managed Control Plane (MCP) servers, improving scalability and deployment flexibility. Builders can integrate custom functions alongside sandbox tools, enabling more tailored and secure operations. Additionally, agents can refresh credentials on the fly without losing their internal state, reducing downtime and the need for manual resets.

Why builders should care

These updates reduce operational friction for anyone orchestrating complex AI agent workflows. Background execution means agents don’t block other processes, which is crucial for automations requiring multiple steps or long-running tasks. Direct MCP server access allows distributed deployments that align better with enterprise infrastructure needs or multi-cloud setups. Mixing custom functions with sandboxed environments opens the door for safer, more fine-grained control over agent capabilities. The credential refresh capability minimizes interruptions, making long-running agents more robust in environments where access tokens expire regularly.

The practical takeaway

For developers and operators building with Gemini API, these features translate into higher reliability and flexibility. Asynchronous background execution enables smoother multitasking and better resource management during deployments. Using remote MCP servers provides control over where and how agents run, reducing latency and centralizing management. Custom function support alongside sandbox tools enhances security while extending functionality. Finally, the seamless credential refresh cuts down on manual maintenance and makes continuous operation possible without downtime.

What to watch next

Next moves could include broader support for more diverse integration scenarios or enhanced security controls around agent access and execution. It will also be important to monitor how enterprises adopt remote MCP server usage to see if it drives more distributed AI agent architectures. Further improvements around state persistence or advanced orchestration features would deepen Gemini API’s practical appeal. Builders relying on Managed Agents should evaluate these new capabilities to streamline work pipelines and minimize operational risks.

AI Quick Briefs Editorial Desk