how to get exit_stack in google adk 1.0.0?

[huang4956]

MCPToolset.from_server() support has been removed from the library in version 1.0.0, how do I get the exit_stack for subsequent session closes?

[Sri-Krishna-V]

You can refer this from the Releases
agent.py file structure with MCP tools are now easier and simpler (now vs before).

I think it is handled inside the BaseToolset.

[huang4956]

thanks @Sri-Krishna-V ,I mean "Using MCP Tools in your own Agent out of adk web", how do I get exit_stack.

[satyaloka93]

Came looking for answers as well, as I set up my multi agent app using https://google.github.io/adk-docs/tools/mcp-tools/#using-mcp-tools-in-your-own-agent-out-of-adk-web. The releases example does not work @Sri-Krishna-V .

[seanzhou1023]

To close the sessions. you can use the Runner.close() which will close all mcp sessions used by the agent that is run by the runner:

https://github.com/google/adk-python/blob/0c4f52699bdb6187651e83f0c787faf3a945f74c/src/google/adk/runners.py#L488

[satyaloka93]

I saw some update to documentation in PR that helped clear things up for me:

#310

Also hadn't been using runner.close() @seanzhou1023 , but the new tool.close(). tool.close() closes one of the mcp stdio threads, but now oddly I get two threads spawned when I load using MCPToolset. Also there's a warning with stdio_client about this, not sure why this is happening, setting it up like the example.