Worked on the pydantic-ai repository to enhance the reliability of the MCP server lifecycle, focusing on robust resource management during server initialization. Addressed a critical bug by refactoring context management logic to ensure proper cleanup and teardown when initialization fails, including cases where exceptions occur during client or client-stream setup. Employed Python’s async programming and context managers to guarantee that resources are released correctly across all error paths. Added comprehensive regression tests to verify error handling and prevent future regressions in the initialization flow. The work emphasized error handling, testing, and lifecycle management to improve overall server stability and maintainability.