10 Examples Why cURL is an Awesome CLI Tool

Whether you're developer, DevOps engineer, SysAdmin, QA or in any other technical role, you're surely familiar with cURL - the command line tool and library for transferring data with URLs (as described in docs).

Most of the time however, we all really only use curl for simple tasks, such downloading a file or checking if website is accessible, yet there's some much more curl can do!

And in this article we will go through exactly those cool examples and tricks to showcase why curl is awesome and underappreciated tool...

Globbing

First up is globbing, allowing us to make multiple requests with a single curl command:


curl -s "https://jsonplaceholder.typicode.com/users/[1-3]" | jq -s .
curl -s "https://jsonplaceholder.typicode.com/users/[0-10:2]" | jq -s .

curl -s "https://jsonplaceholder.typicode.com/photos/{1,6,35}" | jq -s .

curl -s "https://jsonplaceholder.typicode.com/users/[1-3]" -o "file_#1.json"

The first two commands show how we can run a range of requests - first command produces requests to .../users/1, .../users/2, and .../users/3, while the other uses a step option, which gives us 2, 4, 6, 8 and 10. Considering that these particular requests return JSON, we also combine it with jq ... with the -s (slurp) operator, which joins responses of individual requests into single array.

Third examples uses list of specific numbers instead of range - this also works for characters and words - for example, we could use globbing to make requests with multiple protocols: {http,https}://...

The final example combines globbing with output variables, where the #1 variable in filename refers to the range [1-3]. This will produce file_1.json, file_2.json and file_3.json.

Configuration Files

Most of the time, when using curl we probably want to pass in same commandline options, such as proxy settings, request timeouts, headers, etc. That's where curl configuration file named .curlrc might come in handy:


# ~/.curlrc

# some headers
-H "Upgrade-Insecure-Requests: 1"
-H "Accept-Language: en-US,en;q=0.8"

# follow redirects
--location

It's just a text file, where each line represents one option that will be passed to curl. It's read automatically from ~/.curlrc, so you don't need any extra flags, but you can use -K to override or specify different location, e.g.:


curl -K .curlrc https://google.com

Similarly to flags and options, one might also want to pass in credentials. This can be done with --user option, but that will leave the credentials in shell history, so instead we can utilize .netrc file which curl supports:


# ~/.netrc
machine https://authenticationtest.com/HTTPAuth/
login user
password pass

The format includes machine (the URL), login and password - they can be on single line or as shown above, and we can also have multiple of them in single file. To use it, just pass it to curl like:


curl --netrc-file .netrc https://authenticationtest.com/HTTPAuth/

Parallel Requests

We've already talked about ranges of request in globbing section, but what about parallelization? Well, curl can do that too:


curl -I --parallel --parallel-immediate --parallel-max 3 --config websites.txt

curl -I --parallel --parallel-immediate --parallel-max 3 stackoverflow.com google.com example.com

All we need to do, is add --parallel (or -Z) and curl will open up to 50 parallel connections (can be changed with --parallel-max N). Also notice how we're supplying the URLs - first option is --config and a textfile that would look like:


url = "stackoverflow.com"
url = "google.com"
url = "example.com"

Other option is just putting all the URL on commandline. Both of these options also works with non-parallel request!

Formatting & Variables

curl can output a lot of stuff, and sometimes it can be overwhelming, verbose and unnecessary. Luckily, we can use output formatting to only print the things that we're interested in:


curl --silent --output /dev/null --show-error -w @format.txt http://example.com/

# Type: text/html; charset=UTF-8
# Code: 200
#
# From 8.1.0:
# Scheme: http
# Host: example.com
# Port: 80
#
# Read header content (v7.83.0):
# Server: Sat, 29 Jun 2024 13:01:30 GMT

We do so using the -w option and by passing in a format file. To produce output such as above, we can use:


# format.txt
Type: %{content_type}\nCode: %{response_code}\n\n

From 8.1.0:\n\n

Scheme: %{url.scheme}\n
Host: %{url.host}\n
Port: %{url.port}\n

Read header content (v7.83.0):\n
%header{date}

Each variable is enclosed in %{...}. They can be either simple ones like response_code or part of url.<NAME> which refer to URL components, such as host or port. Finally, we can also output response headers with %header{HEADER_NAME} variables.

One of the many nice use cases for formatting is measuring request/response time, which can be done with following format:


# format.txt
     time_namelookup:  %{time_namelookup}s\n
        time_connect:  %{time_connect}s\n
     time_appconnect:  %{time_appconnect}s\n
    time_pretransfer:  %{time_pretransfer}s\n
       time_redirect:  %{time_redirect}s\n
  time_starttransfer:  %{time_starttransfer}s\n
                     ----------\n
          time_total:  %{time_total}s\n

# Output:
     time_namelookup:  0.000765s
        time_connect:  0.111908s
     time_appconnect:  0.000000s
    time_pretransfer:  0.111967s
       time_redirect:  0.000000s
  time_starttransfer:  0.223373s
                     ----------
          time_total:  0.223992s

For full list of variables see docs.

Testing & Troubleshooting

Most common way to use curl is for (network) troubleshooting - oftentimes just making a request to particular URL will provide enough information, but there are more things we can do, for example we can force usage of specific local network interface:


ip link show
# ...
# 3: wlp5s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP mode DORMANT group default qlen 1000

curl --interface wlp5s0 https://example.com

Similarly, specific DNS server can be forced:


curl --dns-ipv4-addr 1.1.1.1 https://example.com

Or we can test for timeout and capture exit code (exit codes):


curl --connect-timeout 30 --silent --output /dev/null \
  --show-error -w 'Total: %{time_total}s\n' http://google.com/ || EXIT_CODE=$?

if [ $EXIT_CODE = 28 ]
then
  echo "Cannot connect (Timeout)."
else
  echo "Can connect."
fi

This can be useful - for example - for testing if proxy server is working (with -x http://proxy.example.com:80).

Trurl

curl is more than just the CLI tool - the project includes libcurl as well as trurl which I want showcase here.

trurl is a dedicated tool for parsing URLs and is a sibling project to curl. It can be installed from source:


sudo apt-get install libcurl4-openssl-dev
git clone https://github.com/curl/trurl.git
cd trurl
make
# ...

And here are couple examples for using it:


trurl --url https://example.com/some/path/to/file.html --get '{path}'
# /some/path/to/file.html

trurl --url "https://example.com/?name=hello" --append query=key=value
# https://example.com/?name=hello&key=value

# Parse as JSON:
./trurl --url "https://example.com/?name=hello" --json
# [
#   {
#     "url": "https://example.com/?name=hello",
#     "parts": {
#       "scheme": "https",
#       "host": "example.com",
#       "path": "/",
#       "query": "name=hello"
#     },
#     "params": [
#       {
#         "key": "name",
#         "value": "hello"
#       }
#     ]
#   }
# ]

First one shows how we can extract URL component, here it's path, but can be e.g. url, scheme, user, password, options or host.

The second one uses the append feature, to add query parameter to the URL.

And final example shows the --json option which outputs the parsed URL as JSON, which is great for further processing.

There's lot more trurl can do - you can check out this video or manual (examples at the bottom).

Sending/Uploading Data

Most of the time we use curl to download or request data, but it can (obviously) also send data. POSTing data with curl is nothing new right?


curl -X POST "https://httpbin.org/post" -H "accept: application/json" --json '{"key": "value"}'

But sending JSON like this, having to alternate the single and double quotes gets annoying pretty quickly, but there's a better way:


jo -p key=value | curl -X POST "https://httpbin.org/post" -H "accept: application/json" --json @-

I think we're all familiar with parsing JSON output of curl with jq, but what about the opposite? - Above we use jo tool which easily creates JSON which we can then pass to curl using the --json option.

Naturally, the --json option can also take input from file, e.g. with --json @data.json.

Protocols

Last but not least here are protocols - usually we would only use HTTP or HTTPS, but curl supports a lot of protocols.

One I want to mention in particular is telnet, which is handy for testing if server listens on specific port, but what if you're on a server/machine where you don't have and can't install telnet? Simply use curl:


# Same as telnet example.com 1234
curl telnet://example.com:1234

Some of the more obscure (funny) protocol options are IMAP, POP3 and SMTP for emails, which means that you can read and send emails with curl. To read them:


curl --url "imaps://imap.gmail.com:993/Inbox;UID=1" --user "user@gmail.com:PASSWORD"

To make this work with GMail specifically, you would need to create app password which is less secure than normal password. If you actually want to try this, do check out also Gmail IMAP docs and these queries for inspiration.

For sending email, you can use:


curl smtp://mail.example.com \
  --mail-from me@example.com \
  --mail-rcpt someone@example.com \
  --upload-file message.txt \
  -u "me@example.com:PASSWORD"

Here the message.txt is the actual email, which needs to follow a specific format, check out this page for examples.

Conclusion

We're at the end, and I'm pretty sure that was at least 10 examples (I stopped counting). But honestly, this really barely scratches the surface - we haven't even touched the libcurl which is the big part of curl.

There's so much more you can do with curl so I recommend exploring both docs and https://everything.curl.dev/.

推荐订阅源

Martin Heinz's Blog