I'm having trouble passing parameters to the MCP tool using agent_instance.send.

[jinwater88]

My mcp tool is as follows:
@mcp.tool() async def create_reminder(time_str: str, content: str,uid: Optional[int] = None) -> dict: result = await reminder_manager.create_reminder(time_str, content,uid) return result

Sending the response using response = await self.agent_instance.send("Please use local tools to process the test parameter") doesn't work because the tool can't receive the UID. Is using result = await actual_agent.call_tool the only reliable method?

[evalstate]

What's the behaviour you are seeing. I'm not sure how from this prompt Please use local tools to process the test parameter the LLM would choose the create_reminder tool? You can use ***CALL_TOOL tool_name {json} in the User message to force a call, but I'd only really recommend using that for testing.

If you want to call the MCP Tool directly bypassing the LLM the call_tool function would be the best way.

[jinwater88]

@evalstate The above message is incorrect. It should be response = await self.agent_instance.send("Remind me to get up at 8 am tomorrow"). However, my time_str: str, content: str, uid: Optional[int] = None has these three parameters. How can I pass these parameters to my tool to create a reminder using self.agent_instance.send? Does the large model directly parse the message to get the parameters and then pass them to the tool?