Google DeepMind · Gemini API
Gemini agents can now
keep running while you sleep.
Managed Agents, unveiled at Google I/O in May, were originally sync-only — good for short tasks, useless for long ones. On July 9, the Gemini API added background execution and MCP support, finally catching up feature-for-feature with Anthropic's and OpenAI's agent platforms.
Catching Up
From sync-only to
peer platform in two months
Managed Agents debuted at Google I/O in May — a managed layer for running agents, handling tool calls, and tracking state, similar in shape to Anthropic's Managed Agents and OpenAI's Assistants. But the initial release was sync-only: the agent ran only while the client stayed connected, so long batch jobs and async pipelines were awkward at best.
On July 9, Gemini API's Managed Agents added background execution and MCP (Model Context Protocol) support. Background execution lets the task continue after the caller disconnects; MCP is the emerging standard for connecting external tools. With both in place, Gemini is finally feature-comparable with the other two major agent platforms.
Two Additions
The two pieces landing today
Independent features on paper — together they close the gap.
Background execution
Fire off a task, disconnect, come back later — the agent keeps running server-side. Long crawls, bulk file processing, batch summarization of conversation logs: work that used to need a home-grown queue now fits inside a single API.
MCP support
Anthropic's open standard for wiring external tools (databases, Git, internal APIs) into an agent, in a common shape. The concrete win: an MCP server you already wrote for Claude or GPT-based agents can be reused from Gemini as-is.
Parity Check
Three platforms, one row
| Gemini Agents until today | Gemini Agents from 7/9 |
|---|---|
| Sync calls only | Background + sync |
| Google-only tool connectors | MCP servers from any vendor |
| Off the shortlist | Peer-tier to Anthropic and OpenAI |
What To Do
Worth putting on the eval
"We looked at Gemini for our agent platform but the sync limit stopped us" is a line that's come up a lot over the last two months. That constraint drops today. Because long-running batch jobs and external tool integration are the two places agent platforms usually get stuck, closing both at once means your comparison spreadsheet can put a real checkmark in the Gemini column now.
No direct impact for consumer Gemini (web/mobile). This is an update for people wiring agent pipelines through the API.