Trouble using GPT-OSS 120B via MCP (repro + logs inside)
Hi,
I'm developing a chatbot using GPT-120B (locally hosted) because it supports tools and the Model Context Protocol (MCP). However, I'm encountering inconsistent behavior even with the simple demo (e.g. the get_weather tool).
Sometimes it runs fine, but other times I get this error:
agents.exceptions.ModelBehaviorError: Tool get_hrs not found in agent Assistant
I've tried multiple frameworks LangGraph, CrewAI, and the OpenAI SDK but the same issue persists. Interestingly, when I switch to models like gpt-4o-mini, everything works perfectly.
I've spent weeks trying to debug this, but I still can't find a solution. Could you please help me understand what might be causing this problem, or if there are any known compatibility issues with GPT-120B and MCP?
By the way Since im using API in my code to connect to gpt-oss-120 api to communicate with my LLM i think that im dont have to use "Harmony Response format right"
Thank you so much for your time and support.