agent retries failed mcp tool calls with same args forever, no backoff

shipped a mcp-based agent last week and discovered that when a tool call fails (network timeout, rate limit, whatever), the agent just retries with the exact same arguments forever. no exponential backoff, no max retry count, no variation in args. tested on autogen 0.4.2 and langchain 0.3.14, same behavior on both. ended up wrapping every tool in custom retry logic with exponential backoff but feels like the frameworks should handle this