nREPL is a Clojure network REPL that provides a REPL server and client, along with some common APIs of use to IDEs and other tools that may need to evaluate Clojure code in remote environments.



nREPL is available in Maven central. Add it to your Maven project's pom.xml:


or your leiningen project.clj:

[org.clojure/tools.nrepl "0.0.5"]

Please note the changelog below.

nREPL is compatible with Clojure 1.1.0 - 1.3.0.

Embedding nREPL

This library is in its infancy. More info to come. In the meantime, check out the tests or for usage examples.


Building nREPL

  1. Clone the repo
  2. Make sure you have maven installed
  3. Run the maven build; run either:
    1. mvn package: This will produce an nREPL jar file in the target directory, and run all tests with the most recently-released build of Clojure (currently 1.2.0). Or,
    2. mvn verify: This does the same, but also runs the tests with other Clojure "profiles" (currently v1.1.0 and v1.1.0 + clojure-contrib).

Need Help?

Ping cemerick on freenode irc or twitter.

Why another REPL implementation?

There are various Clojure REPL implementations: most notably swank-clojure, as well as others associated with various tools and IDEs. So, why write another?

First, while swank-clojure is widely used due to its association with emacs, there is no Clojure swank client implementation that non-emacs tools could use. Further, swank's Common Lisp/SLIME roots mean that its design and future development are not ideal for serving the needs of users of Clojure remote REPLs.

Second, other network REPL implementations are incomplete and/or not suitable for key use cases.

nREPL has been designed in conjunction with the leads of various Clojure development tools, with the aim of ensuring that it satisfies the requirements of both application developers (in support of activities ranging from interactive remote debugging and experimentation in development contexts through to more advanced use cases such as updating deployed applications) as well as toolmakers (providing a standard way to introspect running environments as a way of informing user interfaces of all kinds).

It is hoped that users of emacs/SLIME will also be able to use nREPL, either by extending SLIME itself to work with its protocol, or by implementing a swank-compatible adapter for nREPL.

The network protocol used is simple, depending neither on JVM or Clojure specifics, thereby allowing (encouraging?) the development of non-Clojure REPL clients. The REPLs operational semantics are such that essentially any future non-JVM Clojure implementations should be able to implement it (hopefully within this same project as a separate batch of methods), with allowances for hosts that lack the concurrency primitives to support e.g. asynchronous evaluation, interrupts, etc.

For more information about the motivation, architecture, use cases, and discussion related to nREPL, see the original design notes, available here.


This section will likely only be of use/interest to those looking to build non-Clojure nREPL client implementations.

It is intended that nREPL's protocol and implementation semantics are such that:

  • compatible servers may be implemented for any version/host combination that Clojure is likely to be found on (with reasonable degradation otherwise)
  • non-Clojure and non-JVM clients may be written without undue difficulty


nREPL's protocol is text- and message-oriented, with all communications encoded using UTF-8.

Each message is a set of key/value pairs prefixed by the number of entries in the resulting map, each atom of which is delineated by a linebreak. Keys may only be strings; values may be strings or numbers. A pseudocode formalism for the message "format" might be:

(<string: key>
 (<string: value> | <number: value>)

String keys and values must be quoted and escaped per Clojure standards (which should align well with e.g. json, etc). Certain slots may have multiple entries, represented as multiple key/value pairs with the same key. Such slots should be treated as having set semantics with regard to the pairs' values.

One can connect to a running nREPL server with a telnet client and interact with it using its "wire" message protocol; this isn't intended as a proper usage example, but may be a handy way for toolmakers and REPL client authors to get a quick feel for the protocol and/or do dirty debugging. Example:

[catapult:~] chas% telnet localhost <port-number>
Connected to localhost.
Escape character is '^]'.
"(println 5)"

Websockets / STOMP support

Coming Soon

Interested in helping? Ping cemerick on freenode irc or twitter.


Client Requests

Only one type of request message is defined, which is used for sending code to the server to be loaded/evaluated. All other REPL behaviours (such as environment introspection, symbol completion, quit/restart, etc.) can be accomplished through evaluating code.

Each request message consists of the following slots:

  • id Must be a unique string identifying the request. UUIDs are suitable, and automatic in the provided nREPL client.
  • code The code to be evaluated.
  • in A string containing content to be bound (via a Reader) to *in* for the duration of code's execution
  • timeout (optional) The maximum amount of time, in milliseconds, that the provided code will be allowed to run before a timeout response is sent. If not provided, a default timeout will be assigned by the server (currently always 60s).
  • accept (optional, experimental) The data type(s) that the client would like to receive back as "rendering(s)" of the value of code. Each type is provided as a string, e.g. "jpg" or "txt"; if multiple data types are acceptable, then multiple accept pairs should be sent in the message. Each type should correspond to a different rendering implementation registered with or provided by the nREPL server; if the value obtained from evaluating code has a rendering implementation matching any of the types in accept, a response message will be sent that contains a rendered-XXX slot, where XXX is the type in question. e.g. if code will return a graph data structure, then including "jpg" in accept might result in a rendered-jpg slot message being returned with image data. Including "txt" in accept might result in a rendered-txt slot message being returned with an ASCII art representation of the graph data structure.

Only id and code are required in every request.

Server Responses

The server will produce multiple messages in response to each client request, each of which can have the following slots:

  • id The ID of the request for which the response was generated.
  • ns The stringified value of *ns* at the time of the response message's generation.
  • out Contains content written to *out* while the request's code was being evaluated. Messages containing *out* content may be sent at the discretion of the server, though at minimum corresponding with flushes of the underlying stream/writer.
  • err Same as out, but for *err*.
  • value The result of printing a result of evaluating a form in the code sent in the corresponding request. More than one value may be sent, if more than one form can be read from the request's code string. In contrast to the output written to *out* and *err*, this may be usefully/reliably read and utilized by the client, e.g. in tooling contexts, assuming the evaluated code returns a printable and readable value. Interactive clients will likely want to simply stream value's content to their UI's primary output / log. Values are printed with prn by default; alternatively, if all of the following conditions hold at the time of printing, a pretty-printer will be used instead:
    1. One of the following is available:
      1. Clojure [1.2.0) (and therefore clojure.pprint)
      2. Clojure Contrib (and therefore clojure.contrib.pprint)
    2.*pretty-print* is set!'ed to true (which persists for the duration of the client connection)
  • rendered-<datatype> (optional, experimental) If the corresponding request contained data type(s) in its accept slot, and the value of the evaluated code sent could be rendered into the specified data type, then rendered-<datatype> will contain the Base64-encoded binary representation of that datatype. e.g. if the request's accept slot included "png", and an implementation existed to produce an image rendering of the request's code's value, then a response message will be sent with a rendered-png slot containing Base64-encoded image data. The details of the data returned via messages containing rendered-<datatype> slots is defined solely by the rendering implementations provided by or registered with the nREPL server.
  • status One of:
    • error Indicates an error occurred evaluating the requested code. The related exception is bound to *e per usual, and printed to *err*, which will be delivered via a later message. The caught exception is printed using prn by default; if*print-stack-trace-on-error* is set!'ed to true (which persists for the duration of the client connection), then exception stack traces are automatically printed to *err* instead.
    • timeout Indicates that the timeout specified by the requesting message expired before the code was fully evaluated.
    • interrupted Indicates that the evaluation of the request's code was interrupted.
    • server-failure An unrecoverable error occurred in conjunction with the processing of the request's code. This probably indicates a bug or fatal system fault in the server itself.
    • done Indicates that the request associated with the specified ID has been completely processed, and no further messages related to it will be sent. This does not imply "success", only that a timeout or interrupt condition was not encountered.

Only the id and ns slots will always be defined. Other slots are only defined when new related data is available (e.g. err when new content has been written to *err*, etc).

Note that evaluations that timeout or are interrupted may nevertheless result in multiple response messages being sent prior to the timeout or interrupt occurring.

Timeouts and Interrupts

Each message has a timeout associated with it, which controls the maximum time that a message's code will be allowed to run before being interrupted and a response message being sent indicating a status of timeout.

The processing of a message may be interrupted by a client by sending another message containing code that invokes the function, providing it with the string ID of the message to be interrupted. The interrupt will be responded to separately as with any other message. (The provided client implementation provides a simple abstraction for handling responses that makes issuing interrupts very straightforward.)

Note that interrupts are performed on a “best-effort” basis, and are subject to the limitations of Java’s threading model. For more read here and here.

Change Log


  • added Clojure 1.3.0 (ALPHA) compatibility


  • fixed (hacked) obtaining clojure.test output when clojure.test is initially loaded within an nREPL session
  • eliminated 1-minute default timeout on expression evaluation
  • all standard REPL var bindings are now properly established and maintained within a session


Thanks to the following Clojure masters for their helpful feedback during the initial design phases of nREPL:

  • Justin Balthrop,
  • Meikel Brandmeyer,
  • Hugo Duncan,
  • Christophe Grand,
  • Phil Hagelberg,
  • Rich Hickey,
  • Chris Houser,
  • Laurent Petit, and
  • Eric Thorsen


Copyright © 2010 Chas Emerick

Licensed under the EPL. (See the file epl.html.)

Tip: Filter by directory path e.g. /media app.js to search for public/media/app.js.
Tip: Use camelCasing e.g. ProjME to search for
Tip: Filter by extension type e.g. /repo .js to search for all .js files in the /repo directory.
Tip: Separate your search with spaces e.g. /ssh pom.xml to search for src/ssh/pom.xml.
Tip: Use ↑ and ↓ arrow keys to navigate and return to view the file.
Tip: You can also navigate files with Ctrl+j (next) and Ctrl+k (previous) and view the file with Ctrl+o.
Tip: You can also navigate files with Alt+j (next) and Alt+k (previous) and view the file with Alt+o.