A newly reported issue highlights significant challenges when connecting a Model Context Protocol (MCP) server to the google-adk framework using the MCPToolset integration. Users attempting to run browser automation and structured testing with Playwright and Google Maps tools encounter critical validation errors. Specifically, when initializing the agent stack, the system throws ´pydantic_core._pydantic_core.ValidationError: 2 validation errors for LlmAgent tools.0.callable´, indicating a fundamental mismatch or misconfiguration in expected callable types within the agent definition.
This interoperability failure occurs even when the setup strictly follows official guidance, including using compatible software versions: google-adk 1.2.1, litellm 1.72.2, openai 1.75.0, python-dotenv 1.1.0, and Python 3.11, on MacOS 15.1.1. The developer´s code attempts to wire up an LlmAgent with Playwright and MCPToolset, passing environment keys and parameters as prescribed. However, Pydantic validation in the LlmAgent expects ´tools´ list items to be callable or instances of BaseTool, but receives MCPToolset objects instead—an incompatibility that halts the agent´s instantiation, blocking all downstream automation tasks and orchestration workflows.
Documentation and sample snippets show intended use for orchestrating test plans via a root agent coordinating browser test agents. Despite attempts to use alternate MCP toolsets, reproduce the error under various conditions, and double-check Python environment settings, the core issue persists. A successful resolution would allow seamless orchestration of sub-agents for test plan execution, reporting, and integration with Artificial Intelligence models (e.g., Gemini), yet currently no workaround is provided. The issue remains open, and community support or updates from the google-adk maintainers are actively sought to restore critical functionality for automation practitioners using the MCPToolset module.
