I’ve noticed a difference in behavior between GPT-5.5 and GPT-5.6 when using MCP connectors.
With GPT-5.5, once the MCP connector was used in the initial request, follow-up questions would automatically continue using the same connector and context. However, with GPT-5.6, follow-up questions no longer appear to use the MCP connector reliably it is hit or miss. I have to explicitly tag the connector in every prompt to get it to use it again.
This wasn’t necessary with GPT-5.5, so it seems like there’s been a change in how connector context is handled across follow-up conversations. Could someone confirm if this is expected behavior or if it’s a regression?