### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

Environment

LiteLLM version: (1.86.2) Python: 3.12 Client: OpenAI Codex Desktop 26.602.40724 (wire_api = "responses") Target model: gemini/gemma-4-31b-it via Google AI Studio OS: Windows 11

Bug Description When Codex sends a Responses API request containing tools of type=namespace (used for MCP server namespaces such as mcp__node_repl), LiteLLM converts them to Gemini function_declarations entries that contain only the name field. All description and parameters are silently dropped. The resulting tool declaration is an empty shell that no model can meaningfully invoke, causing every downstream call to return unsupported call. This was verified using the /utils/transform_request endpoint.

Minimal Reproduction ```python import requests, json

payload = { "call_type": "completion", "request_body": { "model": "gemini/gemma-4-31b-it", "tools": [ {"type": "namespace", "name": "mcp__node_repl"} ], "messages": [{"role": "user", "content": "test"}] } }

res = requests.post( "http://127.0.0.…