# Connecting to an NLU server

You can connect a [Rasa NLU-only server](https://legacy-docs-oss.rasa.com/docs/rasa/nlu-only#running-an-nlu-server) to a separately running Rasa dialogue management only server by adding the connection details to the dialogue management server's endpoint configuration file:

## endpoints.yml

```yaml
nlu:
  url: "http://<your nlu host>:<your nlu port>"
  token: <token> # [optional]
  token_name: <name of the token> # [optional] (default: token)
```

The `token` and `token_name` refer to optional [authentication parameters](https://legacy-docs-oss.rasa.com/docs/rasa/http-api#token-based-auth).

The dialogue management server should serve a model that does not include an NLU model. To obtain a dialogue management only model, train a model with `rasa train core` or use `rasa train` but exclude all NLU data.

When the dialogue management server receives a message, it will [send a request](/content/docs/rasa/pages/http-api#operation/parseModelMessage/index.html) to `http://<your nlu host>:<your nlu port>/model/parse` and use the parsing information returned.

### endpoint configuration

The endpoint configuration for the dialogue management server will include an `nlu` endpoint that refers to your NLU only server. Therefore you should **use a separate endpoint configuration file** for the NLU server, excluding the `nlu` endpoint.

If you are implementing a custom NLU server (i.e. not Rasa NLU), your server should provide a `/model/parse` endpoint that responds to requests in the same format as a Rasa NLU server does.