As easy as /aitch-tee-tee-pie/ ūü•ß Modern, user-friendly command-line HTTP client for the API era. JSON support, colors, sessions, downloads, plugins & more.

starsStars 49232
forksForks 3195
watchersWatchers 49232
current-versionCurrent version 2.3.0
total-releasesTotal releases 35
open_issues_countOpen issues 141
dateFirst release 2014-01-25
dateLatest release 2020-10-25
updateLast update 2020-12-23

HTTPie: human-friendly CLI HTTP client for the API era

HTTPie (pronounced aitch-tee-tee-pie) is a command-line HTTP client. Its goal is to make CLI interaction with web services as human-friendly as possible. HTTPie is designed for testing, debugging, and generally interacting with APIs & HTTP servers. The http & https commands allow for creating and sending arbitrary HTTP requests. They use simple and natural syntax and provide formatted and colorized output.


This documentation is best viewed at, where you can select your corresponding HTTPie version as well as run examples directly from the browser using a embedded terminal. If you are reading this on GitHub, then this text covers the current development version. You are invited to submit fixes and improvements to the the docs by editing README.rst.

  • Expressive and intuitive syntax
  • Formatted and colorized terminal output
  • Built-in JSON support
  • Forms and file uploads
  • HTTPS, proxies, and authentication
  • Arbitrary request data
  • Custom headers
  • Persistent sessions
  • Wget-like downloads
  • Linux, macOS and Windows support
  • Plugins
  • Documentation
  • Test coverage

On macOS, HTTPie can be installed via Homebrew (recommended):

$ brew install httpie

A MacPorts port is also available:

$ port install httpie

Most Linux distributions provide a package that can be installed using the system package manager, for example:

# Debian, Ubuntu, etc.
$ apt install httpie
# Fedora
$ dnf install httpie
# CentOS, RHEL, ...
$ yum install httpie
# Gentoo
$ emerge httpie
# Arch Linux
$ pacman -S httpie

A universal installation method (that works on Windows, Mac OS X, Linux, …, and always provides the latest version) is to use pip:

# Make sure we have an up-to-date version of pip and setuptools:
$ python -m pip install --upgrade pip setuptools

$ python -m pip install --upgrade httpie

(If pip installation fails for some reason, you can try easy_install httpie as a fallback.)

Python version 3.6 or greater is required.

You can also install the latest unreleased development version directly from the master branch on GitHub. It is a work-in-progress of a future stable release so the experience might be not as smooth.

On macOS you can install it with Homebrew:

$ brew uninstall --force httpie
$ brew install --HEAD httpie

Otherwise with pip:

$ pip install --upgrade

Verify that now we have the current development version identifier with the -dev suffix, for example:

$ http --version
# 2.0.0-dev

Hello World:

$ https


$ http [flags] [METHOD] URL [ITEM [ITEM]]

See also http --help.

Custom JSON data:

$ http PUT X-API-Token:123 name=John

Submitting forms:

$ http -f POST hello=World

See the request that is being sent using one of the output options:

$ http -v

Build and print a request without sending it using offline mode:

$ http --offline hello=offline

Use GitHub API to post a comment on an issue with authentication:

$ http -a USERNAME POST body='HTTPie is awesome! :heart:'

Upload a file using redirected input:

$ http < files/data.json

Download a file and save it via redirected output:

$ http > image.png

Download a file wget style:

$ http --download

Use named sessions to make certain aspects of the communication persistent between requests to the same host:

$ http --session=logged-in -a username:password API-Key:123
$ http --session=logged-in

Set a custom Host header to work around missing DNS records:

$ http localhost:8000

The name of the HTTP method comes right before the URL argument:

$ http DELETE

Which looks similar to the actual Request-Line that is sent:

DELETE /delete HTTP/1.1

When the METHOD argument is omitted from the command, HTTPie defaults to either GET (with no request data) or POST (with request data).

The only information HTTPie needs to perform a request is a URL.

The default scheme is http:// and can be omitted from the argument:

$ http
# =>

HTTPie also installs an https executable, where the default scheme is https://:

$ https
# =>

If you find yourself manually constructing URLs with querystring parameters on the terminal, you may appreciate the param==value syntax for appending URL parameters.

With that, you don’t have to worry about escaping the & separators for your shell. Additionally, any special characters in the parameter name or value get automatically URL-escaped (as opposed to parameters specified in the full URL, which HTTPie doesn’t modify).

$ http q==httpie per_page==1
GET /search/repositories?q=httpie&per_page=1 HTTP/1.1

Additionally, curl-like shorthand for localhost is supported. This means that, for example :3000 would expand to http://localhost:3000 If the port is omitted, then port 80 is assumed.

$ http :/foo
GET /foo HTTP/1.1
Host: localhost
$ http :3000/bar
GET /bar HTTP/1.1
Host: localhost:3000
$ http :
GET / HTTP/1.1
Host: localhost

When HTTPie is invoked as https then the default scheme is https:// ($ https will make a request to

You can also use the --default-scheme <URL_SCHEME> option to create shortcuts for other protocols than HTTP (possibly supported via plugins). Example for the httpie-unixsocket plugin:

# Before
$ http http+unix://%2Fvar%2Frun%2Fdocker.sock/info
# Create an alias
$ alias http-unix='http --default-scheme="http+unix"'
# Now the scheme can be omitted
$ http-unix %2Fvar%2Frun%2Fdocker.sock/info

The standard behaviour of HTTP clients is to normalize the path portion of URLs by squashing dot segments as a typically filesystem would:

$ http -v
GET /etc/password HTTP/1.1

The --path-as-is option allows you to disable this behavior:

$ http --path-as-is -v
GET /../../etc/password HTTP/1.1

There are a few different request item types that provide a convenient mechanism for specifying HTTP headers, simple JSON and form data, files, and URL parameters.

They are key/value pairs specified after the URL. All have in common that they become part of the actual request that is sent and that their type is distinguished only by the separator used: :, =, :=, ==, @, =@, and :=@. The ones with an @ expect a file path as value.

Item Type Description
HTTP Headers Name:Value Arbitrary HTTP header, e.g. X-API-Token:123.
URL parameters name==value Appends the given name/value pair as a query string parameter to the URL. The == separator is used.
Data Fields field=value, field=@file.txt Request data fields to be serialized as a JSON object (default), to be form-encoded (with --form, -f), or to be serialized as multipart/form-data (with --multipart).
Raw JSON fields field:=json, field:=@file.json Useful when sending JSON and one or more fields need to be a Boolean, Number, nested Object, or an Array, e.g., meals:='["ham","spam"]' or pies:=[1,2,3] (note the quotes).
Fields upload fields field@/dir/file field@file;type=mime Only available with --form, -f and --multipart. For example screenshot@~/Pictures/img.png, or 'cv@cv.txt;type=text/markdown'. With --form, the presence of a file field results in a --multipart request.

Note that data fields aren’t the only way to specify request data: Redirected input is a mechanism for passing arbitrary request data.

You can use \ to escape characters that shouldn’t be used as separators (or parts thereof). For instance, foo\==bar will become a data key/value pair (foo= and bar) instead of a URL parameter.

Often it is necessary to quote the values, e.g. foo='bar baz'.

If any of the field names or headers starts with a minus (e.g., -fieldname), you need to place all such items after the special token -- to prevent confusion with --arguments:

$ http  --  -name-starting-with-dash=foo -Unusual-Header:bar
POST /post HTTP/1.1
-Unusual-Header: bar
Content-Type: application/json

    "-name-starting-with-dash": "foo"

JSON is the lingua franca of modern web services and it is also the implicit content type HTTPie uses by default.

Simple example:

$ http PUT name=John
PUT / HTTP/1.1
Accept: application/json, */*;q=0.5
Accept-Encoding: gzip, deflate
Content-Type: application/json

    "name": "John",
    "email": ""

If your command includes some data request items, they are serialized as a JSON object by default. HTTPie also automatically sets the following headers, both of which can be overwritten:

Content-Type application/json
Accept application/json, */*;q=0.5

You can use --json, -j to explicitly set Accept to application/json regardless of whether you are sending data (it’s a shortcut for setting the header via the usual header notation: http url Accept:'application/json, */*;q=0.5'). Additionally, HTTPie will try to detect JSON responses even when the Content-Type is incorrectly text/plain or unknown.

Non-string JSON fields use the := separator, which allows you to embed arbitrary JSON data into the resulting JSON object. Additionally, text and raw JSON files can also be embedded into fields using =@ and :=@:

$ http PUT \
    name=John \                        # String (default)
    age:=29 \                          # Raw JSON ‚ÄĒ Number
    married:=false \                   # Raw JSON ‚ÄĒ Boolean
    hobbies:='["http", "pies"]' \      # Raw JSON ‚ÄĒ Array
    favorite:='{"tool": "HTTPie"}' \   # Raw JSON ‚ÄĒ Object
    bookmarks:=@files/data.json \      # Embed JSON file
    description=@files/text.txt        # Embed text file
PUT /person/1 HTTP/1.1
Accept: application/json, */*;q=0.5
Content-Type: application/json

    "age": 29,
    "hobbies": [
    "description": "John is a nice guy who likes pies.",
    "married": false,
    "name": "John",
    "favorite": {
        "tool": "HTTPie"
    "bookmarks": {
        "HTTPie": "",

Please note that with the request items data field syntax, commands can quickly become unwieldy when sending complex structures. In such cases, it’s better to pass the full raw JSON data via redirected input, for example:

$ echo '{"hello": "world"}' | http POST
$ http POST < files/data.json

Furthermore, this syntax only allows you to send an object as the JSON document, but not an array, etc. Here, again, the solution is to use redirected input.

Submitting forms is very similar to sending JSON requests. Often the only difference is in adding the --form, -f option, which ensures that data fields are serialized as, and Content-Type is set to, application/x-www-form-urlencoded; charset=utf-8. It is possible to make form data the implicit content type instead of JSON via the config file.

$ http --form POST name='John Smith'
POST /post HTTP/1.1
Content-Type: application/x-www-form-urlencoded; charset=utf-8


If one or more file fields is present, the serialization and content type is multipart/form-data:

$ http -f POST name='John Smith' cv@~/files/data.xml

The request above is the same as if the following HTML form were submitted:

<form enctype="multipart/form-data" method="post" action="">
    <input type="text" name="name" />
    <input type="file" name="cv" />

Please note that @ is used to simulate a file upload form field, whereas =@ just embeds the file content as a regular text field value.

When uploading files, their content type is inferred from the file name. You can manually override the inferred content type:

$ http -f POST name='John Smith' cv@'~/files/data.bin;type=application/pdf'

To perform a multipart/form-data request even without any files, use --multipart instead of --form:

$ http --multipart --offline hello=world
Content-Length: 129
Content-Type: multipart/form-data; boundary=c31279ab254f40aeb06df32b433cbccb

Content-Disposition: form-data; name="hello"


File uploads are always streamed to avoid memory issues with large files.

By default, HTTPie uses a random unique string as the multipart boundary but you can use --boundary to specify a custom string instead:

$ http --form --multipart --boundary=xoxo --offline hello=world
Content-Length: 129
Content-Type: multipart/form-data; boundary=xoxo

Content-Disposition: form-data; name="hello"


If you specify a custom Content-Type header without including the boundary bit, HTTPie will add the boundary value (explicitly specified or auto-generated) to the header automatically:

http --form --multipart --offline hello=world Content-Type:multipart/letter
Content-Length: 129
Content-Type: multipart/letter; boundary=c31279ab254f40aeb06df32b433cbccb

Content-Disposition: form-data; name="hello"


To set custom headers you can use the Header:Value notation:

$ http  User-Agent:Bacon/1.0  'Cookie:valued-visitor=yes;foo=bar'  \
    X-Foo:Bar  Referer:
GET /headers HTTP/1.1
Accept: */*
Accept-Encoding: gzip, deflate
Cookie: valued-visitor=yes;foo=bar
User-Agent: Bacon/1.0
X-Foo: Bar

There are a couple of default headers that HTTPie sets:

GET / HTTP/1.1
Accept: */*
Accept-Encoding: gzip, deflate
User-Agent: HTTPie/<version>
Host: <taken-from-URL>

Any of these can be overwritten and some of them unset (see bellow).

To unset a previously specified header (such a one of the default headers), use Header::

$ http Accept: User-Agent:

To send a header with an empty value, use Header;:

$ http 'Header;'

The --max-headers=n options allows you to control the number of headers HTTPie reads before giving up (the default 0, i.e., there’s no limit).

$ http --max-headers=100

Use --offline to construct HTTP requests without sending them anywhere. With --offline, HTTPie builds a request based on the specified options and arguments, prints it to stdout, and then exits. It works completely offline; no network connection is ever made. This has a number of use cases, including:

Generating API documentation examples that you can copy & paste without sending a request:

$ http --offline POST server.chess/api/games API-Key:ZZZ w=magnus b=hikaru t=180 i=2
$ http --offline MOVE server.chess/api/games/123 API-Key:ZZZ p=b a=R1a3 t=77

Generating raw requests that can be sent with any other client:

# 1. save a raw request to a file:
$ http --offline POST hello=world > request.http
# 2. send it over the wire with, for example, the fantastic netcat tool:
$ nc 80 < request.http

You can also use the --offline mode for debugging and exploring HTTP and HTTPie, and for ‚Äúdry runs‚ÄĚ.

--offline has the side-effect of automatically activating --print=HB, i.e., both the request headers and the body are printed. You can customize the output with the usual output options, with the exception that there is not response to be printed. You can use --offline in combination with all the other options (e.g., --session).

HTTP clients send cookies to the server as regular HTTP headers. That means, HTTPie does not offer any special syntax for specifying cookies ‚ÄĒ the usual Header:Value notation is used:

Send a single cookie:

$ http Cookie:sessionid=foo
GET / HTTP/1.1
Accept: */*
Accept-Encoding: gzip, deflate
Connection: keep-alive
Cookie: sessionid=foo
User-Agent: HTTPie/0.9.9

Send multiple cookies (note the header is quoted to prevent the shell from interpreting the ;):

$ http 'Cookie:sessionid=foo;another-cookie=bar'
GET / HTTP/1.1
Accept: */*
Accept-Encoding: gzip, deflate
Connection: keep-alive
Cookie: sessionid=foo;another-cookie=bar
User-Agent: HTTPie/0.9.9

If you often deal with cookies in your requests, then chances are you’d appreciate the sessions feature.

The currently supported authentication schemes are Basic and Digest (see auth plugins for more). There are two flags that control authentication:

--auth, -a Pass a username:password pair as the argument. Or, if you only specify a username (-a username), you’ll be prompted for the password before the request is sent. To send an empty password, pass username:. The username:password@hostname URL syntax is supported as well (but credentials passed via -a have higher priority).
--auth-type, -A Specify the auth mechanism. Possible values are basic, digest, or the name of any auth plugins you have installed. The default value is basic so it can often be omitted.

$ http -a username:password

$ http -A digest -a username:password

$ http -a username

$ http -a username:

Authentication information from your ~/.netrc file is by default honored as well.

For example:

$ cat ~/.netrc
login httpie
password test
$ http
HTTP/1.1 200 OK

This can be disabled with the --ignore-netrc option:

$ http --ignore-netrc

Additional authentication mechanism can be installed as plugins. They can be found on the Python Package Index. Here’s a few picks:

By default, HTTP redirects are not followed and only the first response is shown:

$ http

To instruct HTTPie to follow the Location header of 30x responses and show the final response instead, use the --follow, -F option:

$ http --follow

If you additionally wish to see the intermediary requests/responses, then use the --all option as well:

$ http --follow --all

To change the default limit of maximum 30 redirects, use the --max-redirects=<limit> option:

$ http --follow --all --max-redirects=2

You can specify proxies to be used through the --proxy argument for each protocol (which is included in the value in case of redirects across protocols):

$ http --proxy=http: --proxy=https:

With Basic authentication:

$ http --proxy=http:http://user:pass@

You can also configure proxies by environment variables ALL_PROXY, HTTP_PROXY and HTTPS_PROXY, and the underlying Requests library will pick them up as well. If you want to disable proxies configured through the environment variables for certain hosts, you can specify them in NO_PROXY.

In your ~/.bash_profile:

export HTTP_PROXY=
export NO_PROXY=localhost,

Usage is the same as for other types of proxies:

$ http --proxy=http:socks5://user:pass@host:port --proxy=https:socks5://user:pass@host:port

To skip the host’s SSL certificate verification, you can pass --verify=no (default is yes):

$ http --verify=no

You can also use --verify=<CA_BUNDLE_PATH> to set a custom CA bundle path:

$ http --verify=/ssl/custom_ca_bundle

To use a client side certificate for the SSL communication, you can pass the path of the cert file with --cert:

$ http --cert=client.pem

If the private key is not contained in the cert file you may pass the path of the key file with --cert-key:

$ http --cert=client.crt --cert-key=client.key

Use the --ssl=<PROTOCOL> option to specify the desired protocol version to use. This will default to SSL v2.3 which will negotiate the highest protocol that both the server and your installation of OpenSSL support. The available protocols are ssl2.3, ssl3, tls1, tls1.1, tls1.2, tls1.3. (The actually available set of protocols may vary depending on your OpenSSL installation.)

# Specify the vulnerable SSL v3 protocol to talk to an outdated server:
$ http --ssl=ssl3

You can specify the available ciphers with --ciphers. It should be a string in the OpenSSL cipher list format.

$ http --ciphers=ECDHE-RSA-AES128-GCM-SHA256

Note: these cipher strings do not change the negotiated version of SSL or TLS, they only affect the list of available cipher suites.

To see the default cipher string, run http --help and see the --ciphers section under SSL.

By default, HTTPie only outputs the final response and the whole response message is printed (headers as well as the body). You can control what should be printed via several options:

--headers, -h Only the response headers are printed.
--body, -b Only the response body is printed.
--verbose, -v Print the whole HTTP exchange (request and response). This option also enables --all (see below).
--print, -p Selects parts of the HTTP exchange.
--quiet, -q Don't print anything to stdout and stderr.

All the other output options are under the hood just shortcuts for the more powerful --print, -p. It accepts a string of characters each of which represents a specific part of the HTTP exchange:

Character Stands for
H request headers
B request body
h response headers
b response body

Print request and response headers:

$ http --print=Hh PUT hello=world

--verbose can often be useful for debugging the request and generating documentation examples:

$ http --verbose PUT hello=world
PUT /put HTTP/1.1
Accept: application/json, */*;q=0.5
Accept-Encoding: gzip, deflate
Content-Type: application/json
User-Agent: HTTPie/0.2.7dev

    "hello": "world"

HTTP/1.1 200 OK
Connection: keep-alive
Content-Length: 477
Content-Type: application/json
Date: Sun, 05 Aug 2012 00:25:23 GMT
Server: gunicorn/0.13.4


--quiet redirects all output that would otherwise go to stdout and stderr (except for error messages) to /dev/null. This doesn’t affect output to a file via --output or --download.

# There will be no output:
$ http --quiet enjoy='the silence'

To see all the HTTP communication, i.e. the final request/response as well as any possible intermediary requests/responses, use the --all option. The intermediary HTTP communication include followed redirects (with --follow), the first unauthorized request when HTTP digest authentication is used (--auth=digest), etc.

# Include all responses that lead to the final one:
$ http --all --follow

The intermediary requests/response are by default formatted according to --print, -p (and its shortcuts described above). If you’d like to change that, use the --history-print, -P option. It takes the same arguments as --print, -p but applies to the intermediary requests only.

# Print the intermediary requests/responses differently than the final one:
$ http -A digest -a foo:bar --all -p Hh -P H

As an optimization, the response body is downloaded from the server only if it’s part of the output. This is similar to performing a HEAD request, except that it applies to any HTTP method you use.

Let’s say that there is an API that returns the whole resource when it is updated, but you are only interested in the response headers to see the status code after an update:

$ http --headers PATCH name='New Name'

Since we are only printing the HTTP headers here, the connection to the server is closed as soon as all the response headers have been received. Therefore, bandwidth and time isn’t wasted downloading the body which you don’t care about. The response headers are downloaded always, even if they are not part of the output

The universal method for passing request data is through redirected stdin (standard input)‚ÄĒpiping.

By default, stdin data is buffered and then with no further processing used as the request body. If you provide Content-Length, then the request body is streamed without buffering. You can also use --chunked to enable streaming via chunked transfer encoding.

There are multiple useful ways to use piping:

Redirect from a file:

$ http PUT X-API-Token:123 < files/data.json

Or the output of another program:

$ grep '401 Unauthorized' /var/log/httpd/error_log | http POST

You can use echo for simple data:

$ echo '{"name": "John"}' | http PATCH X-API-Token:123

You can also use a Bash here string:

$ http <<<'{"name": "John"}'

You can even pipe web services together using HTTPie:

$ http GET | http POST

You can use cat to enter multiline data on the terminal:

$ cat | http POST
$ cat | http POST Content-Type:text/plain
- buy milk
- call parents

On OS X, you can send the contents of the clipboard with pbpaste:

$ pbpaste | http PUT

Passing data through stdin cannot be combined with data fields specified on the command line:

$ echo 'data' | http POST more=data   # This is invalid

To prevent HTTPie from reading stdin data you can use the --ignore-stdin option.

An alternative to redirected stdin is specifying a filename (as @/path/to/file) whose content is used as if it came from stdin.

It has the advantage that the Content-Type header is automatically set to the appropriate value based on the filename extension. For example, the following request sends the verbatim contents of that XML file with Content-Type: application/xml:

$ http PUT @files/data.xml

File uploads are always streamed to avoid memory issues with large files.

You can use the --chunked flag to instruct HTTPie to use Transfer-Encoding: chunked:

$ http --chunked PUT hello=world
$ http --chunked --multipart PUT hello=world foo@files/data.xml
$ http --chunked @files/data.xml
$ cat files/data.xml | http --chunked

HTTPie does several things by default in order to make its terminal output easy to read.

Syntax highlighting is applied to HTTP headers and bodies (where it makes sense). You can choose your preferred color scheme via the --style option if you don’t like the default one. There dozens of styles available, here are just a few special or notable ones:

auto Follows your terminal ANSI color styles. This is the default style used by HTTPie.
default Default styles of the underlying Pygments library. Not actually used by default by HTTPie. You can enable it with --style=default
monokai A popular color scheme. Enable with --style=monokai.
fruity A bold, colorful scheme. Enable with --style=fruity.
… See $ http --help for all the possible --style values.

Also, the following formatting is applied:

  • HTTP headers are sorted by name.
  • JSON data is indented, sorted by keys, and unicode escapes are converted to the characters they represent.

One of these options can be used to control output processing:

--pretty=all Apply both colors and formatting. Default for terminal output.
--pretty=colors Apply colors.
--pretty=format Apply formatting.
--pretty=none Disables output processing. Default for redirected output.

You can control the applied formatting via the --format-options option. The following options are available:

For example, this is how you would disable the default header and JSON key sorting, and specify a custom JSON indent size:

$ http --format-options headers.sort:false,json.sort_keys:false,json.indent:2

This is something you will typically store as one of the default options in your config file. See http --help for all the available formatting options.

There are also two shortcuts that allow you to quickly disable and re-enable sorting-related format options (currently it means JSON keys and headers): --unsorted and --sorted.

Binary data is suppressed for terminal output, which makes it safe to perform requests to URLs that send back binary data. Binary data is suppressed also in redirected, but prettified output. The connection is closed as soon as we know that the response body is binary,

$ http

You will nearly instantly see something like this:

HTTP/1.1 200 OK
Content-Type: application/octet-stream

| NOTE: binary data not shown in terminal |

HTTPie uses a different set of defaults for redirected output than for terminal output. The differences being:

  • Formatting and colors aren‚Äôt applied (unless --pretty is specified).
  • Only the response body is printed (unless one of the output options is set).
  • Also, binary data isn‚Äôt suppressed.

The reason is to make piping HTTPie’s output to another programs and downloading files work with no extra flags. Most of the time, only the raw response body is of an interest when the output is redirected.

Download a file:

$ http > image.png

Download an image of Octocat, resize it using ImageMagick, upload it elsewhere:

$ http | convert - -resize 25% -  | http

Force colorizing and formatting, and show both the request and the response in less pager:

$ http --pretty=all --verbose | less -R

The -R flag tells less to interpret color escape sequences included HTTPie`s output.

You can create a shortcut for invoking HTTPie with colorized and paged output by adding the following to your ~/.bash_profile:

function httpless {
    # `httpless'
    http --pretty=all --print=hb "$@" | less -R;

HTTPie features a download mode in which it acts similarly to wget.

When enabled using the --download, -d flag, response headers are printed to the terminal (stderr), and a progress bar is shown while the response body is being saved to a file.

$ http --download
HTTP/1.1 200 OK
Content-Disposition: attachment; filename=httpie-master.tar.gz
Content-Length: 257336
Content-Type: application/x-gzip

Downloading 251.30 kB to "httpie-master.tar.gz"
Done. 251.30 kB in 2.73862s (91.76 kB/s)

There are three mutually exclusive ways through which HTTPie determines the output filename (with decreasing priority):

  1. You can explicitly provide it via --output, -o. The file gets overwritten if it already exists (or appended to with --continue, -c).
  2. The server may specify the filename in the optional Content-Disposition response header. Any leading dots are stripped from a server-provided filename.
  3. The last resort HTTPie uses is to generate the filename from a combination of the request URL and the response Content-Type. The initial URL is always used as the basis for the generated filename ‚ÄĒ even if there has been one or more redirects.

To prevent data loss by overwriting, HTTPie adds a unique numerical suffix to the filename when necessary (unless specified with --output, -o).

You can also redirect the response body to another program while the response headers and progress are still shown in the terminal:

$ http -d |  tar zxf -

If --output, -o is specified, you can resume a partial download using the --continue, -c option. This only works with servers that support Range requests and 206 Partial Content responses. If the server doesn’t support that, the whole file will simply be downloaded:

$ http -dco

  • The --download option only changes how the response body is treated.
  • You can still set custom headers, use sessions, --verbose, -v, etc.
  • --download always implies --follow (redirects are followed).
  • --download also implies --check-status (error HTTP status will result in a non-zero exist static code).
  • HTTPie exits with status code 1 (error) if the body hasn‚Äôt been fully downloaded.
  • Accept-Encoding cannot be set with --download.

Responses are downloaded and printed in chunks which allows for streaming and large file downloads without using too much memory. However, when colors and formatting is applied, the whole response is buffered and only then processed at once.

You can use the --stream, -S flag to make two things happen:

  1. The output is flushed in much smaller chunks without any buffering, which makes HTTPie behave kind of like tail -f for URLs.
  2. Streaming becomes enabled even when the output is prettified: It will be applied to each line of the response and flushed immediately. This makes it possible to have a nice output for long-lived requests, such as one to the Twitter streaming API.

Prettified streamed response:

$ http --stream

Streamed output by small chunks à la tail -f:

# Send each new line (JSON object) to another URL as soon as it arrives from a streaming API:
$ http --stream | while read line; do echo "$line" | http ; done

By default, every request HTTPie makes is completely independent of any previous ones to the same host.

However, HTTPie also supports persistent sessions via the --session=SESSION_NAME_OR_PATH option. In a session, custom HTTP headers (except for the ones starting with Content- or If-), cookies (manually specified or sent by the server) persist between requests to the same host.

# Create a new session:
$ http --session=./session.json API-Token:123
# Inspect / edit the generated session file:
$ cat session.json
# Re-use the existing session ‚ÄĒ the API-Token header will be set:
$ http --session=./session.json

All session data, including credentials, cookie data, and custom headers are stored in plain text. That means session files can also be created and edited manually in a text editor‚ÄĒthey are regular JSON. It also means that they can be read by anyone who has access to the session file.

You can create one or more named session per host. For example, this is how you can create a new session named user1 for

$ http --session=user1 -a user1:password X-Foo:Bar

From now on, you can refer to the session by its name (user1). When you choose to use the session again, any previously specified authentication or HTTP headers will automatically be set:

$ http --session=user1

To create or reuse a different session, simple specify a different name:

$ http --session=user2 -a user2:password X-Bar:Foo

Named sessions’s data is stored in JSON files inside the sessions subdirectory of the config directory, typically: ~/.config/httpie/sessions/<host>/<name>.json (%APPDATA%\httpie\sessions\<host>\<name>.json on Windows).

If you have executed the above commands on a unix machine, you should be able list the generated sessions files using:

$ ls -l ~/.config/httpie/sessions/

Instead of a name, you can also directly specify a path to a session file. This allows for sessions to be re-used across multiple hosts:

# Create a session:
$ http --session=/tmp/session.json
# Use the session to make a request to another host:
$ http --session=/tmp/session.json
# You can also refer to a previously created named session:
$ http --session=~/.config/httpie/sessions/

When creating anonymous sessions, please remember to always include at least one /, even if the session files is located in the current directory (i.e., --session=./session.json instead of just --session=session.json), otherwise HTTPie assumes a named session instead.

To use an existing session file without updating it from the request/response exchange after it has been created, specify the session name via --session-read-only=SESSION_NAME_OR_PATH instead.

# If the session file doesn’t exist, then it is created:
$ http --session-read-only=./ro-session.json Custom-Header:orig-value
# But it is not updated:
$ http --session-read-only=./ro-session.json Custom-Header:new-value

TL;DR: Cookie storage priority: Server response > Command line request > Session file

To set a cookie within a Session there are three options:

  1. Get a Set-Cookie header in a response from a server
$ http --session=./session.json
  1. Set the cookie name and value through the command line as seen in cookies
$ http --session=./session.json Cookie:foo=bar
  1. Manually set cookie parameters in the json file of the session
    "__meta__": {
    "about": "HTTPie session file",
    "help": "",
    "httpie": "2.2.0-dev"
    "auth": {
        "password": null,
        "type": null,
        "username": null
    "cookies": {
        "foo": {
            "expires": null,
            "path": "/",
            "secure": false,
            "value": "bar"

Cookies will be set in the session file with the priority specified above. For example, a cookie set through the command line will overwrite a cookie of the same name stored in the session file. If the server returns a Set-Cookie header with a cookie of the same name, the returned cookie will overwrite the preexisting cookie.

Expired cookies are never stored. If a cookie in a session file expires, it will be removed before sending a new request. If the server expires an existing cookie, it will also be removed from the session file.

HTTPie uses a simple config.json file. The file doesn’t exist by default but you can create it manually.

To see the exact location for your installation, run http --debug and look for config_dir in the output.

The default location of the configuration file on most platforms is $XDG_CONFIG_HOME/httpie/config.json (defaulting to ~/.config/httpie/config.json).

For backwards compatibility, if the directory ~/.httpie exists, the configuration file there will be used instead.

On Windows, the config file is located at %APPDATA%\httpie\config.json.

The config directory can be changed by setting the $HTTPIE_CONFIG_DIR environment variable:

$ export HTTPIE_CONFIG_DIR=/tmp/httpie
$ http

Currently HTTPie offers a single configurable option:

An Array (by default empty) of default options that should be applied to every invocation of HTTPie.

For instance, you can use this config option to change your default color theme:

$ cat ~/.config/httpie/config.json
    "default_options": [

Even though it is technically possible to include there any of HTTPie’s options, it is not recommended to modify the default behaviour in a way that would break your compatibility with the wider world as that can generate a lot of confusion.

Default options from the config file, or specified any other way, can be unset for a particular invocation via --no-OPTION arguments passed on the command line (e.g., --no-style or --no-session).

When using HTTPie from shell scripts, it can be handy to set the --check-status flag. It instructs HTTPie to exit with an error if the HTTP status is one of 3xx, 4xx, or 5xx. The exit status will be 3 (unless --follow is set), 4, or 5, respectively.


if http --check-status --ignore-stdin --timeout=2.5 HEAD &> /dev/null; then
    echo 'OK!'
    case $? in
        2) echo 'Request timed out!' ;;
        3) echo 'Unexpected HTTP 3xx Redirection!' ;;
        4) echo 'HTTP 4xx Client Error!' ;;
        5) echo 'HTTP 5xx Server Error!' ;;
        6) echo 'Exceeded --max-redirects=<n> redirects!' ;;
        *) echo 'Other Error!' ;;

The default behaviour of automatically reading stdin is typically not desirable during non-interactive invocations. You most likely want to use the --ignore-stdin option to disable it.

It is a common gotcha that without this option HTTPie seemingly hangs. What happens is that when HTTPie is invoked for example from a cron job, stdin is not connected to a terminal. Therefore, rules for redirected input apply, i.e., HTTPie starts to read it expecting that the request body will be passed through. And since there’s no data nor EOF, it will be stuck. So unless you’re piping some data to HTTPie, this flag should be used in scripts.

Also, it might be good to set a connection --timeout limit to prevent your program from hanging if the server never responds.

The syntax of the command arguments closely corresponds to the actual HTTP requests sent over the wire. It has the advantage that it’s easy to remember and read. It is often possible to translate an HTTP request to an HTTPie argument list just by inlining the request elements. For example, compare this HTTP request:

POST /post HTTP/1.1
X-API-Key: 123
User-Agent: Bacon/1.0
Content-Type: application/x-www-form-urlencoded


with the HTTPie command that sends it:

$ http -f POST \
  X-API-Key:123 \
  User-Agent:Bacon/1.0 \
  name=value \

Notice that both the order of elements and the syntax is very similar, and that only a small portion of the command is used to control HTTPie and doesn’t directly correspond to any part of the request (here it’s only -f asking HTTPie to send a form request).

The two modes, --pretty=all (default for terminal) and --pretty=none (default for redirected output), allow for both user-friendly interactive use and usage from scripts, where HTTPie serves as a generic HTTP client.

As HTTPie is still under heavy development, the existing command line syntax and some of the --OPTIONS may change slightly before HTTPie reaches its final version 1.0. All changes are recorded in the change log.

Please use the following support channels:

Under the hood, HTTPie uses these two amazing libraries:

  • Requests ‚ÄĒ Python HTTP library for humans
  • Pygments ‚ÄĒ Python syntax highlighter

HTTPie plays exceptionally well with the following tools:

  • http-prompt ‚ÄĒ interactive shell for HTTPie featuring autocomplete and command syntax highlighting
  • jq ‚ÄĒ CLI JSON processor that works great in conjunction with HTTPie

Helpers to convert from other client tools:

  • CurliPie help convert cURL command line to HTTPie command line.

  • httpcat ‚ÄĒ a lower-level sister utility of HTTPie for constructing raw HTTP requests on the command line.



BSD-3-Clause: LICENSE.

these fine people have contributed.