Tools for Working with URLs and HTTP

Useful tools for working with HTTP organised by HTTP verbs (GET(), POST(), etc). Configuration functions make it easy to control additional request components (authenticate(), add_headers() and so on).

The aim of httr is to provide a wrapper for the curl package, customised to the demands of modern web APIs.

Key features:

  • Functions for the most important http verbs: GET(), HEAD(), PATCH(), PUT(), DELETE() and POST().

  • Automatic connection sharing across requests to the same website (by default, curl handles are managed automatically), cookies are maintained across requests, and a up-to-date root-level SSL certificate store is used.

  • Requests return a standard reponse object that captures the http status line, headers and body, along with other useful information.

    • Response content is available with content() as a raw vector (as = "raw"), a character vector (as = "text"), or parsed into an R object (as = "parsed"), currently for html, xml, json, png and jpeg.

    • You can convert http errors into R errors with stop_for_status().

  • Config functions make it easier to modify the request in common ways: set_cookies(), add_headers(), authenticate(), use_proxy(), verbose(), timeout(), content_type(), accept(), progress().

  • Support for OAuth 1.0 and 2.0 with oauth1.0_token() and oauth2.0_token(). The demo directory has eight OAuth demos: four for 1.0 (twitter, vimeo, withings and yahoo) and four for 2.0 (facebook, github, google, linkedin). OAuth credentials are automatically cached within a project.

httr wouldn't be possible without the hard work of the authors of curl and libcurl. Thanks! httr is inspired by http libraries in other languages, such as Resty, Requests and httparty.

To get the current released version from CRAN:


To get the current development version from github:

# install.packages("devtools")


httr 1.2.1

  • Fix bug with new cache creation code: need to check that cache isn't an empty file.

httr 1.2.0

  • oauth_signature() no longer prepends 'oauth_' to additional parameters. (@jimhester, #373)

  • All print() methods now invisibly return x (#355).

  • DELETE() gains a body parameter (#326).

  • New encode = "raw" allows you to do your own encoding for requests with bodies.

  • New http_type() returns the content/mime type of a request, sans parameters.

  • No longer uses use custom requests for standard POST requests (#356, #357). This has the side-effect of properly following redirects after POST, fixing some login issues (eg hadley/rvest#133).

  • Long deprecated multipart argument to POST(), PUT() and PATCH() has been removed.

  • The cross-session OAuth cache is now created with permission 0600, and should give a better error if it can't be created (#365).

  • New RETRY() function allows you to retry a request multiple times until it succeeds (#353).

  • The default user agent string is now computed once and cached. This is a small performance improvement, but important for local connections (#322, @richfitz).

  • oauth_callback() gains trailing slash for facebook compatibility (#324).

  • progress() gains con argument to control where progress bar is rendered (#359).

  • When use_basic_auth option is used to obtain a token, token refreshes will now use basic authentication too.

  • Suppress unhelpful "No encoding supplied: defaulting to UTF-8." when printing a response (#327).

  • All auto parser functions now have consistent arguments. This fixes problem where ... is pass on to another function (#330).

  • parse_media() can once again parse multiple parameters (#362, #366).

  • Correctly cast config in POST().

  • Fix in readfunction to close connection when done.

httr 1.1.0

  • stop_for_status(), warn_for_status() and (new) message_for_status() replace message argument with new task argument that optionally describes the current task. This allows API wrappers to provide more informative error messages on failure (#277, #302). stop_for_status() and warn_for_status() return the response if there were no errors. This makes them easier to use in pipelines (#278).

  • url_ok() and url_successful() have been deprecated in favour of the more flexible http_error(), which works with urls, responses and integer status codes (#299).

  • oauth1.0_token() gains RSA-SHA1 signature support with the private_key argument (@nathangoulding, #316).

  • oauth2.0_token() throws an error if it fails to get an access token (#250) and gains two new arguments:

    • user_params allows you to pass arbitrary additional parameters to the token access endpoint when acquiring or refreshing a token (@cornf4ke, #312)

    • use_basic_auth allows you to pick use http authentication when getting a token (#310, @grahamrp).

  • oauth_service_token() checks that its arguments are the correct types (#282) and anways returns a request object (#313, @nathangoulding).

  • refresh_oauth2.0() checks for known OAuth2.0 errors and clears the locally cached token in the presense of any (@nathangoulding, #315).

  • httr no longer bundles cacert.pem, and instead it relies on the bundle in openssl. This bundle is only used a last-resort on windows with R <3.2.0.

  • Switch to 'openssl' package for hashing, hmac, signatures, and base64.

  • httr no longer depends on stringr (#285, @jimhester).

  • build_url() collapses vector path with / (#280, @artemklevtsov).

  • content(x) uses xml2 for XML documents and readr for csv and tsv.

  • content(, type = "text") defaults to UTF-8 encoding if not otherwise specified.

  • has_content() correctly tests for the presence/absence of body content (#91).

  • parse_url() correctly parses urls like file:///a/b/c work (#309).

  • progress() returns TRUE to fix for 'progress callback must return boolean' warning (@jeroenooms, #252).

  • upload_file() supports very large files (> 2.5 Gb) (@jeroenooms, #257).

httr 1.0.0

  • httr no longer uses the RCurl package. Instead it uses the curl package, a modern binding to libcurl written by Jeroen Ooms (#172). This should make httr more reliable and prevent the "easy handle already used in multi handle" error. This change shouldn't affect any code that uses httr - all the changes have happened behind the scenes.

  • The oauth_listener can now listen on a custom IP address and port (the previously hardwired ip:port of is now just the default). This permits authentication to work under other settings, such as inside docker containers (which require localhost uses instead). To configure, set the system environmental variables HTTR_LOCALHOST and HTTR_PORT respectively (@cboettig, #211).

  • POST(encode = 'json') now automatically turns length-1 vectors into json scalars. To prevent this automatic "unboxing", wrap the vector in I() (#187).

  • POST(), PUT() and PATCH() now drop NULL body elements. This is convenient and consistent with the behaviour for url query params.

  • cookies argument to handle() is deprecated - cookies are always turned on by default.

  • brew_dr() has been renamed to httr_dr() - that's what it should've been in the first place!

  • content(type = "text") compares encodings in a case-insensitive manner (#209).

  • context(type = "auto") uses a better strategy for text based formats (#209). This should allow the encoding argument to work more reliably.

  • config() now cleans up duplicated options (#213).

  • Uses CURL_CA_BUNDLE environment variable to look for cert bundle on Windows (#223).

  • safe_callback() is deprecated - it's no longer needed with curl.

  • POST() and PUT() now clean up after themselves when uploading a single file (@mtmorgan).

  • proxy() gains an auth argument which allows you to pick the type of http authentication used by the proxy (#216).

  • VERB() gains body and encode arguments so you can generate arbitrary requests with a body.

  • tumblr added as an oauth_endpoint.

httr 0.6.1

  • Correctly parse headers with multiple :, thanks to @mmorgan (#180).

  • In content(), if no type is provided to function or specified in headers, and we can't guess the type from the extension, we now assume that it's application/octet-stream (#181).

  • Throw error if timeout() is less than 1 ms (#175).

  • Improved LinkedIn OAuth demo (#173).

httr 0.6.0

  • New write_stream() allows you to process the response from a server as a stream of raw vectors (#143).

  • Suport for Google OAuth2 service accounts. (#119, thanks to help from @siddharthab).

  • VERB() allows to you use custom http verbs (#169).

  • New handle_reset() to allow you to reset the handle if you get the error "easy handle already used in multi handle" (#112).

  • Uses R6 instead of RC. This makes it possible to extend the OAuth classes from outside of httr (#113).

  • Now only set capath on Windows - system defaults on linux and mac ox seem to be adequate (and in some cases better). I've added a couple of tests to ensure that this continues to work in the future.

  • vignette("api-packages") gains more detailed instructions on setting environment variables, thanks to @jennybc.

  • Add revoke_all() to revoke all stored tokens (if possible) (#77).

  • Fix for OAuth 2 process when using options(httr_oob_default = TRUE) (#126, @WillemPaling).

  • New brew_dr() checks for common problems. Currently checks if your libCurl uses NSS. This is unlikely to work so it gives you some advice on how to fix the problem (thanks to @eddelbuettel for debugging this problem).

  • Content-Type set to title case to avoid errors in servers which do not correctly implement case insensitivity in header names. (#142, #146) thanks to Håkon Malmedal (@hmalmedal) and Jim Hester (@jimhester).

  • Correctly parse http status when it only contains two components (#162).

  • Correctly parse http headers when field name is followed by any amount (including none) of white space.

  • Default "Accepts" header set to application/json, text/xml, application/xml, */*: this should slightly increase the likelihood of getting xml back. application/xml is correctly converted to text before being parsed to XML::xmlParse() (#160).

  • Make it again possible to override the content type set up by POST() when sending data (#140).

  • New safe_callback() function operator that makes R functions safe for use as RCurl callbacks (#144).

  • Added support for passing oauth1 tokens in URL instead of the headers (#145, @bogstag).

  • Default to out-of-band credential exchange when httpuv isn't installed. (#168)

  • new_token() has been removed - this was always an internal function so you should never have been using it. If you were, switch to creating the tokens directly.

  • Deprecate guess_media(), and instead use mime::guess_type() (#148).

httr 0.5

  • You can now save response bodies directly to disk by using the write_disk() config. This is useful if you want to capture large files that don't fit in memory (#44).

  • Default accept header is now "application/json, text/xml, /" - this should encourage servers to send json or xml if they know how.

  • httr_options() allows you to easily filter the options, e.g. httr_options("post")

  • POST() now specifies Curl options more precisely so that Curl know's that you're doing a POST and can respond appropriately to redirects.

  • Preliminary and experimental support for caching with cache_info() and rerequest() (#129). Be aware that this API is likely to change in the future.

  • parse_http_date() parses http dates according RFC2616 spec.

  • Requests now print the time they were made.

  • Mime type application/xml is automatically parsed with ``XML::xmlParse()`. (#128)

  • Now possible to specify both handle and url when making a request.

  • content(type = "text") uses readBin() instead of rawToChar() so that strings with embedded NULLs (e.g. WINDOWS-1252) can be re-encoded to UTF-8.

  • DELETE() now returns body of request (#138).

  • headers() is now a generic with a method for response objects.

  • parse_media() failed to take into account that media types are case-insenstive - this lead to bad re-encoding for content-types like "text/html; Charset=UTF-8"

  • Typo which broke set_cookies() fixed by @hrbrmstr.

  • url_ok() works correctly now, instead of always returning FALSE, a bug since version 0.4 (#133).

  • Remove redundant arguments simplifyDataFrame and simplifyMatrix for json parser.

httr 0.4

  • New headers() and cookies() functions to extract headers and cookies from responses. Previoulsy internal status_code() function now exported to extract status_code() from responses.

  • POST(), PUT(), and PATCH() now use encode argument to determine how list inputs are encoded. Valid values are "multiple", "form" or "json". The multipart argument is now deprecated (#103). You can stream a single file from disk with upload_file("path/"). The mime type will be guessed from the extension, or can be supplied explicitly as the second argument to upload_file().

  • progress() will display a progress bar, useful if you're doing large uploads or downloads (#17).

  • verbose() now uses a custom debug function so that you can see exactly what data is sent to the server. Arguments control exactly what is included, and the defaults have been selected to be more helpful for the most common cases (#102).

  • with_verbose() makes it easier to see verbose information when http requests are made within other functions (#87).

  • New quickstart vignette to help you get up and running with httr.

  • New api-packages vignette describes how best practices to follow when writing R packages that wrap web APIs.

  • httr_options() lists all known config options, translating between their short R names and the full libcurl names. The curl_doc() helper function allows you to jump directly to the online documentation for an option.

  • authenticate() now defaults to type = "basic" which is pretty much the only type of authentication anyone uses.

  • Updated cacert.pem to version at 2014-04-22 (#114).

  • content_type(), content_type_xml() and content_type_json() make it easier to set the content type for POST requests (and other requests with a body).

  • has_content() tells you if request has any content associated with it (#91).

  • Add is_interactive() parameter to oauth_listener(), init_oauth1.0() and init_oauth2.0() (#90).

  • oauth_signature() and oauth_header() now exported to make it easier to construct custom authentication for APIs that use only some components of the full OAuth process (e.g. 2 legged OAuth).

  • NULL query parameters are now dropped automatically.

  • When print()ing a response, httr will only attempt to print the first few lines if it's a text format (i.e. either the main type is text or is application/json). It will also truncate each line so that it fits on screen - this should hopefully make it easier to see a little bit of the content, without filling the screen with gibberish.

  • new_bin() has been removed: it's easier to see what's going on in examples with

  • user_agent() once again overrides default (closes #97)

  • parse(type = "auto") returns NULL if no content associated with request (#91).

  • Better strategy for resetting Curl handles prevents carry-over of error status and other problems (#112).

  • set_config() and with_config() now work with tokens (#111).

httr 0.3

OAuth 2.0 has recieved a major overhaul in this version. The authentication dance now works in more environments (including RStudio), and is generally a little faster. When working on a remote server, or if R's internet connection is constrained in other ways, you can now use out-of-band authentication, copying and pasting from any browser to your R session. OAuth tokens from endpoints that regularly expire access tokens can now be refreshed, and will be refresh automatically on authentication failure.

httr now uses project (working directory) based caching: every time you create or refresh a token, a copy of the credentials will be saved in .httr-oauth. You can override this default for individual tokens with the cache parameter, or globally with the httr_oauth_cache option. Supply either a logical vector (TRUE = always cache, FALSE = never cache, NA = ask), or a string (the path to the cache file).

You should NOT include this cache file in source code control - if you do, delete it, and reset your access token through the corresponding web interface. To help, httr will automatically add appropriate entries to .gitignore and .Rbuildignore.

These changes mean that you should only ever have to authenticate once per project, and you can authenticate from any environment in which you can run R. A big thanks go to Craig Citro (@craigcitro) from google, who contributed much code and many ideas to make this possible.

  • The OAuth token objects are now reference classes, which mean they can be updated in place, such as when an access token expires and needs to be refreshed. You can manually refresh by calling $refresh() on the object. You can force reinitialisation (to do the complete dance from scratch) by calling $reinit(force = TRUE).

  • If a signed OAuth2 request fails with a 401 and the credentials have a refresh_token, then the OAuth token will be automatically refreshed (#74).

  • OAuth tokens are cached locally in a file called .httr-oauth (unless you opt out). This file should not be included in source code control, and httr will automatically add to .gitignore and .Rbuildignore. The caching policy is described in more detail in the help for the Token class.

  • The OAuth2 dance can now be performed without running a local webserver (#33, thanks to @craigcitro). To make that the default, set options(httr_oob_default = TRUE). This is useful when running R remotely.

  • Add support for passing oauth2 tokens in headers instead of the URL, and make this the default (#34, thanks to @craigcitro).

  • OAuth endpoints can store arbitrary extra urls.

  • Use the httpuv webserver for the OAuth dance instead of the built-in httpd server (#32, thanks to @jdeboer). This makes the dance work in Rstudio, and also seems a little faster. Rook is no longer required.

  • oauth_endpoints() includes some popular OAuth endpoints.

  • HTTP verbs (GET(), POST() etc) now pass unnamed arguments to config() and named arguments to modify_url() (#81).

  • The placement of ... in POST(), PATCH() and PUT() has been tweaked so that you must always specify body and multipart arguments with their full name. This has always been recommended practice; now it is enforced.

  • httr includes its own copy of cacert.pem, which is more recent than the version included in RCurl (#67).

  • Added default user agent which includes versions of Curl, RCurl and httr.

  • Switched to jsonlite from rjson.

  • Content parsers no longer load packages on to search path.

  • stop_for_status() now raises errors with useful classes so that you can use tryCatch() to take different actions depending on the type of error. See http_condition() for more details.

  • httr now imports the methods package so that it works when called with Rscript.

  • New automatic parsers for mime types text/tab-separated-values and text/csv (#49)

  • Add support for fragment in url building/parsing (#70, thanks to @craigcitro).

  • You can suppress the body entirely in POST(), PATCH() and PUT() with body = FALSE.

  • If you supply multiple headers of the same name, the value of the most recently set header will always be used.

  • Urls with missing query param values (e.g. are now parsed correctly (#27). The names of query params are now also escaped and unescaped correctly when parsing and building urls.

  • Default html parser is now XML::htmlParse() which is easier to use with xpath (#66).

httr 0.2

  • OAuth now uses custom escaping function which is guaranteed to work on all platforms (Fixes #21)

  • When concatenating configs, concatenate all the headers. (Fixes #19)

  • export hmac_sha1 since so many authentication protocols need this

  • content will automatically guess what type of output (parsed, text or raw) based on the content-type header. It also automatically converts text content to UTF-8 (using the charset in the media type) and can guess at mime type from extension if server doesn't supply one. Media type and encoding can be overridden with the type and encoding arguments respectively.

  • response objects automatically print content type to aid debugging.

  • text_content has become context(, "text") and parsed_content content(, "parsed"). The previous calls are deprecated and will be removed in a future version.

  • In oauth_listener, use existing httpd port if help server has already been started. This allows the ouath authentication dance to work if you're in RStudio. (Fixes #15).

  • add several functions related to checking the status of an http request. Those are : status, url_ok and url_success as well as stop_for_status and warn_for_status.

  • build_url: correctly add params back into full url.

httr 0.1.1

  • Add new default config: use the standard SSL certificate

  • Add recommendation to use custom handles with authenticate

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.


1.2.1 by Hadley Wickham, 8 months ago

Browse source code at

Authors: Hadley Wickham [aut, cre], RStudio [cph]

Documentation:   PDF Manual  

Task views: Web Technologies and Services

MIT + file LICENSE license

Imports jsonlite, mime, curl, openssl, R6

Suggests httpuv, jpeg, knitr, png, testthat, readr, xml2, rmarkdown

Imported by ALA4R, AntWeb, AutoDeskR, BIOMASS, DrillR, HIBPwned, IMFData, LendingClub, Luminescence, OECD, OpenML, Quandl, RBMRB, RCriteo, RDocumentation, RDota2, REDCapR, RForcecom, RGA, RNeXML, RNeo4j, ROpenFIGI, ROpenWeatherMap, ROptimizely, RPresto, RPublica, RSelenium, RSiteCatalyst, RSocrata, RStripe, RYandexTranslate, RZabbix, RefManageR, SWMPr, SocialMediaLab, SocialMediaMineR, SpaDES, TCGAretriever, WHO, WaterML, WikiSocio, WikidataQueryServiceR, WikidataR, WikipediR, WufooR, aRxiv, abbyyR, abjutils, algorithmia, alm, analogsea, archivist, archivist.github, atsd, aws.alexa, aws.polly,, aws.sns, aws.sqs, backblazer, bea.R, benchmarkme, bigrquery, binman, bioimagetools, biomartr, blscrapeR, bold, boxr, brranching, cablecuttr, camsRad, ccafs, cdcfluview, censys, chillR, chromer, ckanr, covr, cranlogs, crunch, curlconverter, darksky, dataRetrieval, dataonderivatives, dataone, datarobot, dbhydroR, ddeploy, devtools, diezeit, discgolf, distcomp, dpmr, ecb, ecoengine, elastic, elasticsearchr, elevatr, epidata, estatapi, etseed, europepmc, eurostat, federalregister, finreportr, fitbitScraper, fitcoach, flora, foghorn, fulltext, gdns, gender, genderizeR, geojsonio, geojsonlint, geoknife, geoparser, geosapi, gesis, getlandsat, gfer, gistr, githubinstall, gitlabr, gmailr, gmum.r, googleAuthR, googleCloudStorageR, googleformr, googlesheets, gsheet, gym, hansard, haploR, hdr, hoardeR, htmltab, httpcache, httptest, hypothesisr, icpsrdata, idbr, imfr, imguR, internetarchive, junr, knitcitations, kntnr, kokudosuuchi, languagelayeR, lintr, longurl, lucr, lumendb, machina, manifestoR, mapr, miniCRAN, mnis, monkeylearn, move, mregions, mscstexta4r, mscsweblm4r, neotoma, notifyme, oai, opencage, opencpu, originr, osmplotr, owmr, packagetrackr, pageviews, pangaear, pdfetch, pdftables, pewdata, pinnacle.API, pkgcopier, plotly, pollstR, primerTree, prism, pushoverr, pvsR, pxweb, qiitr, qualtRics, quickmapr, rWBclimate, randNames, rapiclient, rbhl, rbison, rclimateca, rclinicaltrials, rcrossref, rdian, rdnb, rdpla, rdrop2, rdryad, rebird, rechonest, redcapAPI, rentrez, repmis, rerddap, rfigshare, rfishbase, rfisheries, rfml, rfoaas, rgbif, rgeoapi, rgeolocate, rgeospatialquality, rgho, ridigbio, riem, rinat, ritis, rnbn, rnoaa, rnpn, rnrfa, robotstxt, rollbar, ropenaq, rorcid, rosetteApi, rotl, rpinterest, rplos, rrefine, rscopus, rsnps, rstatscn, rsunlight, rtimes, rtweet, rtypeform, rusda, rvertnet, rvest, rwars, rwunderground, sbtools, scholar, slackr, smapr, sofa, solr, solrium, solvebio, spareserver, sparklyr, spocc, stmBrowser, stplanr, stringgaussnet, swirl, taxize, telegram, threewords, tidyverse, tigris, traits, transcribeR, translateR, treebase, trelloR, tuber, twitteR, ubeR, udapi, urlshorteneR, vegdata, vembedr, virustotal, vkR, wbstats, webchem, whoami, whoapi, wikipediatrend, xtractomatic, yhatr, yummlyr.

Depended on by CytobankAPI, GAR, LightningR, RFc, RGoogleAnalytics, RGoogleAnalyticsPremium, RGoogleFit, RSauceLabs, Rcolombos, Rfacebook, Rga4gh, Rlinkedin, SkyWatchr, SmarterPoland, TMDb, WikipediaR, birdnik, censusr, couchDB, epxToR, hkex.api, httping, instaR, markmyassignment, ores, osi, rHealthDataGov, request, seleniumPipes, tumblR, ungeneanno.

Suggested by CopulaDTA, HARtools, Rcereal, aemo, bigQueryR, bio3d, dynamichazard, fauxpas, fbRanks, finch, highcharter, htmltidy, jsonlite, leafletR, nat, neurobase, oasis, packrat, soilDB, stubthat, viridis, xml2.

See at CRAN