### Check for existing issues

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

### What happened?

I have several MCP servers registered behind the LiteLLM MCP Gateway. One of them also exposes a UI resource (MCP Apps extension `_meta.ui.resourceUri` on a tool, fetched via `resources/read`). As soon as that server is aggregated in the same session as any other MCP server, calling `resources/read` for its UI resource fails with a `400`, so the UI never renders (silent fallback to plain text/JSON).

With only that one server allowed in the session, `resources/read` works fine, it only breaks once a second server is added.

Expected: the gateway resolves the resource to the server it actually belongs to (its URI is unambiguous) and returns it, the same way it already resolves prompts across multiple aggregated servers.

### Steps to Reproduce

1. Register 2 MCP servers behind the LiteLLM proxy — one of them exposing a resource (e.g. `ui://my-server/widget.html`). 2. Open a client session that allows both servers, e.g. `x-mcp-servers: my-server,other-server`. 3. Call `resources/read` with `uri: "ui://my-server/widget.html"`. 4. Observe the 400 e…