Google DeepMind added four features to gemini-api-managed-agents" class="entity-chip">Gemini API Managed Agents on July 28, 2026. Background execution and direct MCP server connections target developers building long-running, stateful agent workflows.
Key facts
- Four new features added to Gemini API Managed Agents
- Background execution removes HTTP connection requirement
- MCP protocol crossed 13,000+ servers in June 2026
- Google Cloud launched its own MCP server on July 8, 2026
- Credentials refreshable without losing sandbox state
Google DeepMind rolled out four capabilities to Managed Agents in the Gemini API, according to The Decoder. The additions — background execution, remote MCP server connectivity, custom functions alongside sandbox tools, and credential refresh without state loss — are all available through the Gemini Interactions API.
Key Takeaways
- Google DeepMind added background execution and MCP support to Gemini API Managed Agents.
- Four new features target developers building long-running, stateful agent workflows.
Background execution cuts HTTP dependency
The headline feature lets agents run asynchronously in the background with no open HTTP connection required. This removes the need for developers to maintain persistent connections while agents process long-running tasks like document analysis or multi-step data pipelines. Code examples for JavaScript, Python, and cURL are in the documentation.
MCP support bridges to external data
Managed Agents can now connect directly to remote MCP servers, giving them access to internal databases or APIs through Anthropic's open standard. The timing coincides with MCP's rapid adoption: the protocol crossed 13,000+ servers in June 2026, per the knowledge graph, and was backed by Anthropic, OpenAI, and Block under the Linux Foundation on July 8. Google Cloud launched its own MCP server the same day, and AWS followed on July 7.
Custom functions and credential hygiene
The third feature lets developers use custom functions alongside the built-in sandbox tools, blending Google's native capabilities with user-defined logic. The fourth addresses a persistent pain point: credentials like tokens can be refreshed between interactions without losing the sandbox state, preserving agent context across sessions.
Competitive positioning
The update directly competes with Anthropic's Claude Code, which uses MCP extensively (73 sources in the knowledge graph point to this relationship). Google's move to embed MCP support natively in Managed Agents signals that the protocol is becoming table stakes for agent platforms, not just Anthropic's differentiator.
What to watch
Watch for adoption metrics from Google Cloud Next (expected October 2026) and whether Managed Agents usage crosses the milestone of 100,000 active agent deployments within 90 days. Also track whether OpenAI's Agents SDK adds similar MCP-native support.
Source: the-decoder.com
.png)
