### What happened?

A Responses API request that explicitly asks for MCP tools via the gateway (`server_url: litellm_proxy/...`) but resolves **zero** tools — because the API key/team has no access to the MCP server (`allow_all_keys=false` and no object-permission grant), the server name doesn't exist, or `allowed_tools` matches nothing — is silently sent to the model **with no tools at all**.

The model then answers from thin air. The request looks healthy end to end: HTTP 200, and a `list_mcp_tools` spend log with status **success** and an empty response. In our production deployment this cost several days of debugging, because the failure reads as a model-quality problem ("why is it hallucinating?") instead of a permissions problem.

Real capture (main @ 999637883c, unregistered server name):

``` HTTP/1.1 200 OK ... "output":[{"type":"message","content":[{"text":"It seems I can't access external tools or links, including Linktree. You can check the number of links on your Linktree by logging into your account ..."}]}] ```

Expected: a 400 naming the requested server URLs and the likely causes. (Confirmed with the LiteLLM team on Slack that throwing by default is the preferred s…