Google ADK Python v2.11.0
Google ADK Python v2.11.0 introduces graceful execution cancellation, workflow tool confirmation prompts, ModelConsultTool with call budgets, a local SQLite memory service, and opt-in MCP SDK 2.x support.
Why this signal matters
Google ADK Python v2.11.0 brings several workflow and runtime enhancements. Execution runs can now be gracefully stopped by passing an abort signal to Runner, Workflow, and nodes, with /run sse automatically canceling executions when clients disconnect. Tool nodes inside workflows now support human-in-the-loop validation by pausing for user approval via RequestInput rather than failing downstream. The new ModelConsultTool allows agents to consult alternative models during tasks subject to turn- and session-based budgets. For persistence, a local SQLite memory service is accessible via the sqlite:// URI scheme. Additionally, opt-in support for MCP SDK 2.x was added. A breaking change affects Dev UI customization: runtime-config.json is served directly per request over HTTP instead of written to the installed package path, requiring configuration via server flags ( --logo-text , --logo-image-url ).
Actionable summary
ADK Python v2.11.0 adds abort signals to Runner, Workflow, and Nodes; tool confirmation via RequestInput in workflows; ModelConsultTool for inter-model consultations; and SQLite-backed memory. It introduces a breaking change where Dev UI runtime config is dynamically served by the server rather than written directly to package files.
- Agent usefulness
- 94/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Update any workflows relying on package-level edits of Dev UI runtime-config.json to use --logo-text and --logo-image-url server arguments.
- Adopt abort signal handling in Runner , Workflow , or nodes if client-side disconnection handling is needed.
- Opt in to the modern protocol path if migrating to MCP SDK 2.x.