Gemini API agents gain background tasks,

Google added four capabilities to Gemini API Managed Agents on July 7, aimed squarely at production agent workloads rather than demos. The headline feature is async background execution: passing a background: true parameter returns a pollable task ID instead of blocking on a long-running agent job, which matters for anything that takes longer than a typical request timeout.

The API also gained direct remote MCP server connections, letting a Gemini-based agent talk to a Model Context Protocol server without a proxy layer in between, previously a common workaround for connecting Gemini agents to existing MCP tooling. Rounding out the update: custom function calling can now run alongside Google’s built-in sandbox tools in the same agent, and credential and API-key rotation no longer forces the agent to lose its sandbox state mid-session.

Each of these addresses a specific pain point teams run into once an agent moves from prototype to production: long-running tasks that don’t fit a request-response cycle, tool integrations that shouldn’t need custom proxy code, and credential rotation that shouldn’t mean restarting a session.

If you’re running Gemini-based agents against MCP servers today through a homegrown proxy, the direct connection support is worth switching to, it removes a layer of infrastructure you were probably maintaining just to bridge the two.