### Check for existing issues

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

### What happened?

After restarting LiteLLM, MCP servers that already exist in the database disappear from the MCP Servers UI.

The UI is empty because the backend endpoint `GET /v1/mcp/server` returns `[]` with `200 OK`, even though the MCP server rows are still present in Postgres.

If I then create a new MCP server from the UI, the old DB-backed MCP servers and the newly created one all appear again immediately. This makes it look like an MCP runtime/in-memory registry reload is missing on startup, but is triggered by later MCP write operations.

Tested on July 9, 2026.

### Steps to Reproduce

### Environment

- LiteLLM image: `ghcr.io/berriai/litellm:v1.91.1` - Deployment: Docker Compose - Platform: `linux/amd64` - Database: Postgres - LiteLLM started with `--config=/app/config.yaml --port=4000`

### Reproduction Steps

1. Start LiteLLM with MCP servers already stored in Postgres. 2. Open the MCP Servers UI. 3. Confirm the servers are visible. 4. Restart the LiteLLM service/container. 5. Open the MCP Servers UI again. 6. Observe that the UI is empty. 7. Chec…