Google expands Gemini API managed agents with background tasks and remote MCP support

Google has expanded Managed Agents in the Gemini API with a set of features aimed at making agentic applications easier to run outside short, single-request demos. In a July 7 announcement, Google DeepMind developer relations engineer Philipp Schmid and product manager Mariano Cocirio said the update adds background execution, remote Model Context Protocol server integration, custom function calling and credential refresh across interactions.

The change matters because long-running AI agent work often does not fit neatly inside a live HTTP request. Google says developers can now start an interaction asynchronously, receive an identifier immediately, and then poll, stream progress or reconnect while the managed agent continues running on the server. That gives teams a more practical pattern for workflows such as repository analysis, multi-step research, data preparation or other tasks where an agent may need minutes rather than seconds.

What changed for developers

Managed Agents run through the Gemini Interactions API. According to Google, a single endpoint can coordinate model reasoning, code execution, package installation, file handling and web information inside an isolated cloud sandbox. The newly announced remote MCP support is designed to connect those cloud-hosted agents to external tools and internal systems without forcing developers to build custom proxy middleware for every private database or API.

• Background execution lets agents continue work asynchronously while client applications monitor status.
• Remote MCP server integration allows managed agents to call external tool servers from the sandboxed environment.
• Custom function calling can hand selected steps back to the client when local business logic is required.
• Credential refresh lets developers update network configuration for an existing environment while preserving filesystem state, installed packages and cloned repositories.

The announcement is incremental rather than a new model launch, but it reflects a broader shift in AI tooling: cloud APIs are becoming orchestration platforms for persistent software agents. For developers, the key test will be whether managed execution reduces the fragile glue code that normally surrounds agents, including job queues, reconnection logic, tool routing and secret rotation. Google is positioning the Gemini API update as a way to move those responsibilities into a managed runtime while still allowing applications to keep control of private tools and local functions.

The company says developers can explore the Gemini Interactions API overview and managed agents quickstart for custom agent definitions, environment settings, network rules and streaming patterns. As with any agent system that can reach external tools, teams will still need to review permissions, logging and failure handling before connecting production data sources.

Sources

• blog.google

Cover photo by luis gomes on Pexels, used under the Pexels License.