Alphabet expands Gemini API “managed agents,” adding background task support and remote MCP connections

Alphabet’s Google is rolling out new capabilities in its Gemini API Managed Agents offering, aiming to make it easier for developers to deploy agent-style applications that behave more consistently outside of demos. In a post on Google’s developer and innovation blog, the company said it is “expanding Managed Agents in Gemini API” with additional features that support more production workflows. The announcement frames Managed Agents as a managed, developer-facing way to run agent logic on Google’s side rather than having every application recreate orchestration from scratch. Google did not spell out all implementation details in the announcement text, but it specifically cited “background tasks” as a new capability. In practical terms, background tasks are designed to let an agent continue work after the initial request lifecycle, which is typically necessary for longer-running or multi-step operations such as research, data transformation, or tool-driven execution. The company also highlighted “remote MCP” in the Managed Agents update. MCP, or Model Context Protocol, is an open approach for connecting AI systems to external tools and data sources through a standard interface. By calling out remote MCP, Google is indicating that developers can integrate external MCP servers without having to tightly couple everything into a single runtime. Beyond these two items, Google said there are “more” additions, describing the update as a set of new capabilities in Managed Agents that are meant to make agents “reliable” and “production-ready.” The post’s emphasis suggests a focus on operational behavior, including how agents coordinate multi-step work and invoke tools, rather than only improving raw model responses. For Alphabet, the Managed Agents push sits within a broader effort to shift AI development from simple prompt-and-response calls toward agentic systems that can plan, call tools, and iterate. Google has previously been positioning its Gemini ecosystem as an “agent-first” platform, and this latest update continues that theme by putting more of the agent lifecycle into an API construct. What Google did not disclose in the announcement is just as notable. The post did not provide a full changelog of features, limits, pricing changes, regional availability, or a detailed migration guide for existing developers. It also did not provide specific performance benchmarks, uptime or reliability guarantees, or any developer metrics that would allow users to compare the prior managed-agent behavior with the new background task and remote MCP support. For developers building on Gemini API today, the practical next step is likely to check documentation and example code for Managed Agents to understand how background tasks are represented in calls, and how remote MCP endpoints are configured. Google’s developers and AI updates in the weeks following an API feature announcement typically provide the deeper technical specifications required to integrate the changes into production systems.