Hey, this is a known limitation of the BYOK path for thinking models like Kimi and DeepSeek. After the first tool call, reasoning_content gets dropped from the history, and the provider Moonshot rejects the next request.
The same symptom is discussed here: Kimi/kimi-k2.5 cannot be called to execute tool calls in cursor and here for DeepSeek with the same root cause: Compatibility with DeepSeek model's design to return reasoning content after tool calls?.
We’re tracking the issue, but I can’t share an ETA for a fix yet. I’m going to close this thread as a duplicate. For updates, please follow the DeepSeek thread since it’s more active.
Workaround for now: if you can, turn off thinking mode in Moonshot for this model, or use a non-reasoning model. Tool calls should work fine then.