Gemini API Adds MCP and Background Execution: What Changes for Agent Builders
Agentic AIGoogle Deepmind just shipped two updates to Gemini API Managed Agents that remove the most common reasons to reach for a workaround when building production agents: persistent HTTP connections and database access without an intermediary proxy.
What Happened
The Managed Agents API now supports background execution—agents run asynchronously without requiring an open HTTP connection. This changes the operational model for long-running agent tasks: you no longer need to hold a connection open, handle reconnects, or build streaming infrastructure to track completion. Developers trigger execution and retrieve results independently.
The second addition is remote MCP server connectivity. Agents can now connect directly to internal databases and APIs via remote MCP servers, without routing through an intermediary layer. Custom functions are also now usable alongside built-in sandbox tools, and credentials can refresh between interactions without losing sandbox state. All features are accessible through the Gemini Interactions API with code examples in JavaScript, Python, and cURL.
Why It Matters
Background execution and direct MCP access address the two structural weaknesses that pushed agent builders toward self-hosted orchestration over managed APIs. Open HTTP connections made Managed Agents impractical for tasks longer than a few seconds—a timeout waiting to happen. Remote MCP without an intermediary closes the gap between a managed API and a raw model call: you get the session management and built-in tools of a managed agent without needing to expose internal systems through a proxy layer.
The combination is meaningful for enterprise use cases. Long-running agents—research tasks, document processing pipelines, multi-step data retrieval—previously required custom orchestration to avoid the connection management problem. Remote MCP means those same agents can now reach internal databases and private APIs through the same standard interface teams are already building against for Claude and Cursor integrations.
What This Means for Builders
Teams currently running orchestration frameworks like LangGraph or Temporal specifically because Gemini’s managed layer didn’t support long-running tasks or native database access should re-evaluate the trade-off. For simple to mid-complexity agent workflows, Gemini’s Managed Agents API is now meaningfully more competitive. The growing MCP ecosystem makes this a particularly useful signal: a managed agent layer that speaks MCP natively connects directly to the tooling teams have already built.
The So What
Google’s updates signal a consolidation pattern forming across major AI API providers: managed agent layers are absorbing the infrastructure primitives that once required custom orchestration. Background execution, credential rotation, and MCP connectivity are becoming commodity API features rather than differentiators. Builders who deferred managed APIs because they lacked these primitives should take a fresh look—the table is filling in faster than expected.
Source: The Decoder
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →