Hi all,
I’m in the process of integrating a custom MCP server with Deep Research models and have run into a frustrating limitation. For my use case, it’s essential to pass additional parameters to the search function in order to prefilter results effectively.
However, to use MCP with DR it currently restricts the function signature to search(query: str), which means I’m forced to build auxiliary tools or intermediate steps to set these parameters separately. Even with that, the LLM often bypasses those steps and goes straight to invoking the search function, ignoring the prefilters entirely.
It would be immensely helpful if the search tool could support additional keyword arguments, or at the very least accept **kwargs to allow for more flexible integration patterns.
Is anyone else dealing with this, or are there workarounds you’ve found effective? Would love to hear thoughts and whether this is on the radar for improvement.
Thanks!