Building Servers

This page is very incomplete and a work in progress.


This part of the documentation aims to guide you in the process of implementing an nREPL server. The focus is going to be on implementing the language-agnostic nREPL protocol, not on specifics of the references nREPL server implementation for Clojure.


To build a server compatible with the nREPL protocol all you need to do is build a solution that handles the protocol ops and uses bencode as the default data exchange format.

You can completely ignore concepts like middleware, handles and so on - those are just implementation details from the references implementation. If you want to leverage them that’d be great, but if you don’t - it’s not a big deal.

It’s important to understand that clients only care whether some op is there, that it behaves as expected, and that the data format is the right one. The implementation details are transparent to them.

The basic requirements for a server would be:

Obviously most clients would work fairly well if messages like interrupt are missing, but your users will probably appreciate having those.

Some clients (e.g. CIDER) might be relying on the server startup message to extract connection information from it. That’s why it’s a good to provide the same message for your own server implementation. Here’s the message in question:

nREPL server started on port 58567 on host - nrepl://

Here are a few simple nREPL implementations that you can peruse for inspiration:

It might also be a good idea to see the message exchange between some nREPL server and client. CIDER provides a convenient solution for this.

Best Practices

  • Write an .nrepl-port file on server startup.

  • Add support for .nrepl.edn (at least if you’re working with a language than can easily handle EDN).

  • Avoid adding runtime dependencies, unless you can somehow isolate them. You don’t want your server to mess up dependencies in the user space.

  • Provide some information in the describe op that makes it easy for clients to recognize your server.

  • Provide streaming results to the clients (e.g. values and output), as an option. This generally means that results are split over several messages, so that clients would get feedback faster.

  • Provide some means for the clients to configure (pretty-)printing of results.

Extending the Core Protocol

cider-nrepl defines a bunch of ops that are supported by many Clojure nREPL clients (e.g. info and complete). You can add those ops to your server implementation as well.