Model is not calling tools anymore from MCP
# support
b
call id: 019dd486-0821-7000-a45b-290d1993529c 019dd3da-5755-7007-b61e-668f28c1c9fd The model is loaded with MCP tools within the system prompt (confirmed from the logs) but it never calls the tools and hallucinates top to bottom. Suddenly happening since this morning. Need to have an urgent look into this since this is breaking our product.
Also the frontend for call logs > assistant tab is broken
a
Hey, I’ve dealt with this exact MCP/tool-calling issue before. When the model suddenly stops calling tools and starts hallucinating, it’s usually a routing or tool schema parsing issue (sometimes triggered by a silent API/model update). I can quickly debug the call flow, inspect your system prompt + tool definitions, and fix the triggering logic so tools get called properly again. @big_doggo_best
s
Hi there, thanks for reaching out.
We looked into those call IDs and found that while the tools are being sent correctly, the GPT-4.1 model is simply choosing not to trigger them. This is a known issue with GPT-4.1's consistency with tool calls.
To fix this immediately: Switch your model: Move to GPT-5 variants. These are much more reliable for tool calling right now. Update your prompt: Add a strict instruction like: "You MUST use the provided tools for every request. Do not answer from memory." Switching models should resolve the hallucinations and restore your product's functionality immediately.
Let us know if you still face any issues after updating the model.