generated using jug version

Hello World!


jug() %>%
  get("/", function(req, res, err){
    "Hello World!"
  }) %>%
  simple_error_handler_json() %>%
Serving the jug at

What is jug?

jug is a small web development framework for R which relies heavily upon the httpuv package. It’s main focus is to make building APIs for your code as easy as possible.

jug is not supposed to be either an especially performant nor an uber stable web framework. Other tools (and languages) might be more suited for that. It’s main focus is to easily allow you to create APIs for your R code. However, the flexibility of jug means that, in theory, you could built an extensive web framework with it.

Getting started

To install the latest version use devtools:


Or install the CRAN version:


Load the library:


The jug instance

Everything starts with a jug instance. This instance is created by simply calling jug():

## A Jug instance with 0 middlewares attached

jug is made to work closely with the piping functionality of magrittr (%>%). The configuration of the jug instance is set up by piping the instance through the various functions explained below.


In terms of middleware, jug somewhat follows the specification of middleware by Express. In jug, middleware is a function with access to the request (req), response (res) and error (err) object.

Multiple middlewares can be defined. The order in which the middlewares are added matters. A request will start with being passed through the first middleware added (more specifically the functions specified in it - see next paragraph). It will continue to be passed through the added middlewares until a middleware does not return NULL (note: if a value is set using e.g. res$json("foo") the body will not be NULL). Whatever will be passed by that middleware will be set as the response body.

Most middleware will accept a func or ... argument to which respectively a function or multiple functions can be passed. If multiple functions are passed; the order in which they are passed will be respected when processing a request. To each function the req, res and err objects will be passed (and they thus should accept them).

Method insensitive middleware

The use function is a method insensitive middleware specifier. While it is method insensitive, it can be bound to a specific path. If the path argument (accepts a regex string with grepl setting perl=TRUE) is set to NULL it also becomes path insensitive and will process every request.

A path insensitive example:

jug() %>%
  use(path = NULL, function(req, res, err){
    "test 1,2,3!"
    }) %>%
$ curl
test 1,2,3!

The same example, but path sensitive:

jug() %>%
  use(path = "/", function(req, res, err){
    "test 1,2,3!"
    }) %>%
$ curl
curl: (52) Empty reply from server

$ curl
test 1,2,3!

It is however possible to specify a method to bind to using use (check out ?use), this way you can process request methods for which no prespecified middlewares exist.

Note that in the above example errors / missing route handling is missing (the server might crash / not respond), more on that later.

Method sensitive middleware

In the same style as the request method insensitive middleware, there is request method sensitive middleware available. More specifically, you can use the get, post, put and delete functions.

This type of middleware is bound to a path using the path argument. If path is set to NULL it will bind to every request to the path, given that it is of the corresponding request method.

jug() %>%
  get(path = "/", function(req, res, err){
    "get test 1,2,3!"
    }) %>%
$ curl
get test 1,2,3!

Middlewares are meant to be chained, so to bind different functions to different paths:

jug() %>%
  get(path = "/", function(req, res, err){
    "get test 1,2,3 on path /"
    }) %>%
  get(path = "/my_path", function(req, res, err){
    "get test 1,2,3 on path /my_path"
    }) %>%
$ curl
get test 1,2,3 on path /

$ curl
get test 1,2,3 on path /my_path

Websocket protocol

By default all middleware convenience function bind to the http protocol. You can however access the jug server through websocket by using the websocket sensitive middleware function ws. Below an example echo’ing the incoming message.

jug() %>%
   ws("/echo_message", function(binary, message, res, err){
  }) %>%

Opening a connection to ws:// and sending e.g. the message test to it will then return the value test.

Please note that websocket support is experimental at this stage.

Including elsewhere defined middleware chains

In order to make you code more modular, you can include elsewhere defined middleware chains into your jug instance. To do this you can use a combination of the collector() and include() functions.

Below a collector is defined locally (in the same R script) and included.

    collector() %>%
    get("/", function(req,res,err){

  res<-jug() %>%
    include(collected_mw) %>%

However, it is also possible to include a collector that is defined in another .R file.

Let’s say below is the file my_middlewares.R:


  collector() %>%
  get("/", function(req,res,err){

We can include it as follows:

res<-jug() %>%
  include(collected_mw, "my_middlewares.R") %>%

Predefined middleware

Error handling

A simple error handling middleware (simple_error_handler / simple_error_handler_json) which catches unbound paths and func evaluation errors. If you do not implement a custom error handler, I suggest you add either of these to your jug instance. The simple_error_handler returns an HTML error page while the simple_error_handler_json returns a JSON message.

$ curl
<!DOCTYPE html>
<html lang="en">
    <meta charset="utf-8">
    <title>Not found</title>
    <p>No handler bound to path</p>

If you want to implement your own custom error handling just have a look at the code of these simple error handling middlewares.

Please note that generally you would like the error handler middleware to be attached to the jug instance after all other middleware has been specified.

Easily using your own functions

The main reason jug was created is to easily allow access to your own custom R functions. The convenience function decorate is built especially for this purpose.

If you decorate your own function it will translate all arguments passed in the query string of the request as arguments to your function. It will also pass all headers to the function as arguments.

If your function does not accept a ... argument, all query/header parameters that are not explicitly requested by your function are dropped. If your function requests a req, res or err argument (or ...) the corresponding objects will be passed.


jug() %>%
  get("/", decorate(say_hello)) %>%

If in the above, you pass a parameter name through either the query string or as a header in the GET request, it will return as in the example below.

$ curl
hello Bart !

Static file server

The serve_static_file middleware allows for serving static files.

The default root directory is the one returned by getwd() but can be specified by providing a root_path argument to the serve_static_files middleware. It transforms a bare / path to index.html.

Aside from development, I do not recommend using jug to serve static files.

CORS functionality

CORS functionality is introduced by the cors() middleware function.

Consider the following example.

jug() %>%
  cors() %>%
  get("/", function(req, res, err){
    "Hello World!"
  }) %>%
$ curl -v
*   Trying
* Connected to ( port 8080 (#0)
> GET / HTTP/1.1
> Host:
> User-Agent: curl/7.43.0
> Accept: */*
< HTTP/1.1 200 OK
< Content-Type: text/html
< Access-Control-Allow-Origin: *
< Access-Control-Allow-Methods: POST,GET,PUT,OPTIONS,DELETE,PATCH
< Content-Length: 12
* Connection #0 to host left intact

As you see this adds some default CORS-headers. Check out ?cors for the configuration options, note that you can also add CORS headers to a specific path by specifying the path parameter.


Currently there is only built-in support for basic authentication (check: through the auth_basic middleware function. The middleware will check the request for a valid username / password combination. If an invalid combination is passed, it will return a 401 status, a WWW-Authenticate header and a text body which states that there was an authentication error.

First you will need to define a function that accepts username and password arguments. The funtion should return TRUE if the combination is valid and FALSE if the combination is invalid. A dummy example is shown below. Note, that this function could also check e.g. a database to validate the combo.

# dummy account checker
account_checker <- function(username, password){
  # do something to verify the username and password and return TRUE if combination OK
  all(username == "test_user", 
      password == "test_password")

Next you need to instantiate the auth_basic middleware in you middleware chain. The auth_basic function accepts as first parameter the username/password validation function. Below two examples are given. The first one shows how to do authentication for a specific path (/test).

jug() %>%
  get("/", function(req, res, err){
    "/ req"
  }) %>%
  get("/test", auth_basic(account_checker), function(req, res, err){
    "/test req"
  }) %>%

The second example below shows how to activate basic authentication for all paths in the jug instance.

jug() %>%
  use(NULL, auth_basic(account_checker)) %>%
  get("/", function(req, res, err){
    "/ req"
  }) %>%

Event listeners

The concept of event listeners have been available since version As middelwares did not suffice to implement a robust logger, the concept of events and event listeners was introduced. Currently listeners can bind to events, an example is given below:

jug() %>%
  get("/", function(req,res,err){"foo"}) %>%
  on("finish", function(req, res, err){
    print("the finish event was received; request processing finished!")}
    ) %>%

Three events are available currently:

  • start: this event is triggered once a new request is received
  • finish: this event is triggered once a request has been fully processed
  • error: this event is triggered once an error is being raised inside a middleware

The start and finish event will pass the current state of the req, res and err objects to the listener functions. The error event will pass a fourth argument, namely a character representation of the error message.

Predefined event listeners


A logger based on futile.logger is available through logger.

An example is given below:

jug() %>%
  get("/", function(req,res,err){"foo"}) %>%
  get("/err", function(req,res,err){stop("bar")}) %>%
  logger(threshold = futile.logger::DEBUG, log_file='logfile.log', console=TRUE) %>%
  simple_error_handler_json() %>%

In the above example, the logger threshold is set at futile.logger::DEBUG meaning that we will receive detailed information during the execution. A visual distinction will be made based on the type of message send, e.g. INFO (on path /) or ERROR (on path /err).

In this example, the logger will write to logfile.log and will output to the console.

For more information on the logger thresholds, have a look at the documentation of the futile.logger package.

The request, response and error objects

Request (req) object

The req object contains the request specifications. It has different attributes:

  • req$params a named list of the parameters passed by either the query string, a JSON body, URL parameters or a multipart form
  • req$path the request path
  • req$method the request method
  • req$raw the raw request object as passsed by httpuv
  • req$body the full request body as a character string
  • req$protocol either http or websocket
  • req$headers a named list of the headers in the request (as lowercase and stripped from the HTTP_ prefix provided by the underlying httpuv framework)

It has the following functions attached to it:

  • req$get_header(key) returns the value associated to the specified key in the request (no need to worry about the HTTP_ prefix)
  • req$set_header(key, value) allows to set / alter a header while processing the request (can be useful to pass data to the next middleware)
  • req$attach(key, value) attach a variable to req$params

Response (res) object

The res object contains the response specifications. It has different attributes:

  • res$headers a named list of the set headers
  • res$status the status of the response (defaults to 200)
  • res$body the body of the response (is automatically set to be the content of the not NULL returning middleware or by methods such as res$json())

It also has a set of functions:

  • res$set_header(key, value) set a custom header
  • res$content_type(type) set your own content type (MIME)
  • res$set_status(status) set the status of the response
  • res$text(body) to explicitely set the body of the response
  • res$json(obj, auto_unbox=TRUE) converts an object to JSON, sets it as the body and set the correct content type
  • res$plot(plot_obj, base64=TRUE) convenience function to return a plot object as the response body, the returned plot can either be a base64 representation of the image (default) or the actual binary data

Error (err) object

The err object contains a list of errors, accessible through err$errrors. You can add an error to this list by calling err$set(error). The error will be converted to a character.

Refer to the “Error handling” paragraph for more details.

URL dispatching

The path parameter in the get, post, … functions are processed as being regex patterns.

If there are named capture groups in the path definition, they will be attached to the req$params object. For example the pattern /test/(?<id>.*)/(?<id2>.*) will result in the variables id and id2 (with their respective values) being bound to the req$params object.

If a path pattern is not started with a start of string ^ regex token or ended with an end of string token $, these will be explicitely inserted at respectively the beginning and end of the path pattern specification. For example the path pattern / will be converted to ^/$.

Starting the jug instance

Simply call serve_it() at the end of your piping chain (see Hello World! example).

Practical examples

Minimal CRUD TODO app

A minimal TODO app built in Angular with a jug backend.

Clone the repository to check it out:

Exposing a machine learning model

Let’s train (in a very simplistic way) a linear regression model on the mtcars dataset and assume that our objective is to predict the miles per gallon or mpg variable based on the inputs gear and hp.

##                    mpg cyl disp  hp drat    wt  qsec vs am gear carb
## Mazda RX4         21.0   6  160 110 3.90 2.620 16.46  0  1    4    4
## Mazda RX4 Wag     21.0   6  160 110 3.90 2.875 17.02  0  1    4    4
## Datsun 710        22.8   4  108  93 3.85 2.320 18.61  1  1    4    1
## Hornet 4 Drive    21.4   6  258 110 3.08 3.215 19.44  1  0    3    1
## Hornet Sportabout 18.7   8  360 175 3.15 3.440 17.02  0  0    3    2
## Valiant           18.1   6  225 105 2.76 3.460 20.22  1  0    3    1
  lm(mpg~gear+hp, data=mtcars)

## Call:
## lm(formula = mpg ~ gear + hp, data = mtcars)
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -4.7977 -2.4288 -0.7685  2.2405  7.5943 
## Coefficients:
##              Estimate Std. Error t value Pr(>|t|)    
## (Intercept) 17.755144   3.241809   5.477 6.74e-06 ***
## gear         3.176520   0.762584   4.165 0.000255 ***
## hp          -0.063931   0.008206  -7.791 1.36e-08 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## Residual standard error: 3.108 on 29 degrees of freedom
## Multiple R-squared:  0.7513, Adjusted R-squared:  0.7341 
## F-statistic: 43.79 on 2 and 29 DF,  p-value: 1.731e-09

As we went through a lot of hard work to end up with this model (/s), we now want to expose it through an API. This way we allow other people or applications to make predictions using this model.

As a first step we need to build a minimal prediction function.

predict_mpg <- function(gear, hp){
          newdata = data.frame(gear=as.numeric(gear), 

We can test the function by supplying the gear and hp arguments.

predict_mpg(gear = 4, hp = 80)
## [1] 25.34671

Now, to expose this function as a web API, we need to build a jug instance. We can use the built-in decorate middleware to ease the integration of the predict_mpg function. Below, a minimal example is shown.

jug() %>%
  post("/predict-mpg", decorate(predict_mpg)) %>%
  simple_error_handler_json() %>%
Serving the jug at

We can now send a http POST request to the url and it will return the predicted value! It works out of the box with either the parameters in a JSON body, as multipart/form-data or as a x-www-form-urlencoded.

JSON body

curl -X POST \ \
  -H 'content-type: application/json' \
  -d '{"hp": 80, "gear": 4}'

multipart form

curl -X POST \ \
  -H 'content-type: multipart/form-data; boundary=----WebKitFormBoundary7MA4YWxkTrZu0gW' \
  -F hp=80 \
  -F gear=4

urlencode form

curl -X POST \ \
  -H 'content-type: application/x-www-form-urlencoded' \
  -d 'gear=4&hp=80'