Worked on the pydantic/pydantic-ai repository to enhance error handling within the MCP integration by converting McpError exceptions from MCP tool calls into ModelRetry exceptions. This approach preserved the original error messages, ensuring that downstream processes received clear and actionable information while preventing cascading failures in production workflows. Developed and integrated comprehensive unit tests to verify both the conversion and re-raising behavior, emphasizing reliability and traceability. Utilized Python as the primary language, focusing on robust error handling and thorough unit testing practices. The work addressed a critical bug, improving the resilience and maintainability of the MCP integration without introducing new features.