### Check for existing issues

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

### What happened?

**Summary**

When mcp_semantic_tool_filter is enabled, LiteLLM forwards MCP tools to OpenAI-compatible backends (e.g. LM Studio) in an invalid schema.

Tools are sent without the required function wrapper, causing strict OpenAI-compatible APIs to return 400 BadRequest.

**Environment**

- LiteLLM Proxy (v1.83.10) - Backend: LM Studio (OpenAI-compatible API) - Model: openai/gpt-oss-20b - Feature: mcp_semantic_tool_filter: enabled

**Actual Behavior**

LiteLLM sends tools in this format:

```json "tools": [ { "type": "function", "name": "mymcp-tracker_get_issue", "description": "...", "parameters": { ... } } ] ```

Backend error:

```json [ { "code": "invalid_type", "expected": "object", "received": "undefined", "path": [0, "function"], "message": "Required" } ] ```

**Expected Behavior**

Tools should follow OpenAI function-calling schema:

```json "tools": [ { "type": "function", "function": { "name": "mymcp-tracker_get_issue", "description": "...", "parameters": { ... } …