Package eventsource implements a client and server to allow streaming data one-way over a HTTP connection using the Server-Sent Events API http://dev.w3.org/html5/eventsource/ The client and server respect the Last-Event-ID header.
Package httphead contains utils for parsing HTTP and HTTP-grammar compatible text protocols headers.
Package linkheader provides functions for parsing HTTP Link headers
Package secureheader adds some HTTP header fields widely considered to improve safety of HTTP requests.
Package httpin helps decoding an HTTP request to a custom struct by binding data with querystring (query params), HTTP headers, form data, JSON/XML payloads, URL path params, and file uploads (multipart/form-data).
Package accept allows for easy handling of HTTP Accept headers.
Package httpreadat provides an io.ReaderAt for http requests using the Range header.
Package cloudflarebp provides a round tripper to not get detected by CloudFlare directly on the first HTTP request The round tripper will add required/validated request headers and updates the client TLS configuration It'll NOT solve challenges provided by CloudFlare, just prevent from being detected on the first request
Package websocket implements the WebSocket protocol defined in RFC 6455. The Conn type represents a WebSocket connection. A server application calls the Upgrader.Upgrade method from an HTTP request handler to get a *Conn: Call the connection's WriteMessage and ReadMessage methods to send and receive messages as a slice of bytes. This snippet of code shows how to echo messages using these methods: In above snippet of code, p is a []byte and messageType is an int with value websocket.BinaryMessage or websocket.TextMessage. An application can also send and receive messages using the io.WriteCloser and io.Reader interfaces. To send a message, call the connection NextWriter method to get an io.WriteCloser, write the message to the writer and close the writer when done. To receive a message, call the connection NextReader method to get an io.Reader and read until io.EOF is returned. This snippet shows how to echo messages using the NextWriter and NextReader methods: The WebSocket protocol distinguishes between text and binary data messages. Text messages are interpreted as UTF-8 encoded text. The interpretation of binary messages is left to the application. This package uses the TextMessage and BinaryMessage integer constants to identify the two data message types. The ReadMessage and NextReader methods return the type of the received message. The messageType argument to the WriteMessage and NextWriter methods specifies the type of a sent message. It is the application's responsibility to ensure that text messages are valid UTF-8 encoded text. The WebSocket protocol defines three types of control messages: close, ping and pong. Call the connection WriteControl, WriteMessage or NextWriter methods to send a control message to the peer. Connections handle received close messages by calling the handler function set with the SetCloseHandler method and by returning a *CloseError from the NextReader, ReadMessage or the message Read method. The default close handler sends a close message to the peer. Connections handle received ping messages by calling the handler function set with the SetPingHandler method. The default ping handler sends a pong message to the peer. Connections handle received pong messages by calling the handler function set with the SetPongHandler method. The default pong handler does nothing. If an application sends ping messages, then the application should set a pong handler to receive the corresponding pong. The control message handler functions are called from the NextReader, ReadMessage and message reader Read methods. The default close and ping handlers can block these methods for a short time when the handler writes to the connection. The application must read the connection to process close, ping and pong messages sent from the peer. If the application is not otherwise interested in messages from the peer, then the application should start a goroutine to read and discard messages from the peer. A simple example is: Connections support one concurrent reader and one concurrent writer. Applications are responsible for ensuring that no more than one goroutine calls the write methods (NextWriter, SetWriteDeadline, WriteMessage, WriteJSON, EnableWriteCompression, SetCompressionLevel) concurrently and that no more than one goroutine calls the read methods (NextReader, SetReadDeadline, ReadMessage, ReadJSON, SetPongHandler, SetPingHandler) concurrently. The Close and WriteControl methods can be called concurrently with all other methods. Web browsers allow Javascript applications to open a WebSocket connection to any host. It's up to the server to enforce an origin policy using the Origin request header sent by the browser. The Upgrader calls the function specified in the CheckOrigin field to check the origin. If the CheckOrigin function returns false, then the Upgrade method fails the WebSocket handshake with HTTP status 403. If the CheckOrigin field is nil, then the Upgrader uses a safe default: fail the handshake if the Origin request header is present and the Origin host is not equal to the Host request header. The deprecated package-level Upgrade function does not perform origin checking. The application is responsible for checking the Origin header before calling the Upgrade function. Connections buffer network input and output to reduce the number of system calls when reading or writing messages. Write buffers are also used for constructing WebSocket frames. See RFC 6455, Section 5 for a discussion of message framing. A WebSocket frame header is written to the network each time a write buffer is flushed to the network. Decreasing the size of the write buffer can increase the amount of framing overhead on the connection. The buffer sizes in bytes are specified by the ReadBufferSize and WriteBufferSize fields in the Dialer and Upgrader. The Dialer uses a default size of 4096 when a buffer size field is set to zero. The Upgrader reuses buffers created by the HTTP server when a buffer size field is set to zero. The HTTP server buffers have a size of 4096 at the time of this writing. The buffer sizes do not limit the size of a message that can be read or written by a connection. Buffers are held for the lifetime of the connection by default. If the Dialer or Upgrader WriteBufferPool field is set, then a connection holds the write buffer only when writing a message. Applications should tune the buffer sizes to balance memory use and performance. Increasing the buffer size uses more memory, but can reduce the number of system calls to read or write the network. In the case of writing, increasing the buffer size can reduce the number of frame headers written to the network. Some guidelines for setting buffer parameters are: Limit the buffer sizes to the maximum expected message size. Buffers larger than the largest message do not provide any benefit. Depending on the distribution of message sizes, setting the buffer size to a value less than the maximum expected message size can greatly reduce memory use with a small impact on performance. Here's an example: If 99% of the messages are smaller than 256 bytes and the maximum message size is 512 bytes, then a buffer size of 256 bytes will result in 1.01 more system calls than a buffer size of 512 bytes. The memory savings is 50%. A write buffer pool is useful when the application has a modest number writes over a large number of connections. when buffers are pooled, a larger buffer size has a reduced impact on total memory use and has the benefit of reducing system calls and frame overhead. Per message compression extensions (RFC 7692) are experimentally supported by this package in a limited capacity. Setting the EnableCompression option to true in Dialer or Upgrader will attempt to negotiate per message deflate support. If compression was successfully negotiated with the connection's peer, any message received in compressed form will be automatically decompressed. All Read methods will return uncompressed bytes. Per message compression of messages written to a connection can be enabled or disabled by calling the corresponding Conn method: Currently this package does not support compression with "context takeover". This means that messages must be compressed and decompressed in isolation, without retaining sliding window or dictionary state across messages. For more details refer to RFC 7692. Use of compression is experimental and may result in decreased performance.
Package grab provides a HTTP download manager implementation. Get is the most simple way to download a file: Get will download the given URL and save it to the given destination directory. The destination filename will be determined automatically by grab using Content-Disposition headers returned by the remote server, or by inspecting the requested URL path. An empty destination string or "." means the transfer will be stored in the current working directory. If a destination file already exists, grab will assume it is a complete or partially complete download of the requested file. If the remote server supports resuming interrupted downloads, grab will resume downloading from the end of the partial file. If the server does not support resumed downloads, the file will be retransferred in its entirety. If the file is already complete, grab will return successfully. For control over the HTTP client, destination path, auto-resume, checksum validation and other settings, create a Client: You can monitor the progress of downloads while they are transferring:
Package secheaders checks the HTTP security response headers of a URL and reports which recommended headers are missing or weak.
Package accept allows for easy handling of HTTP Accept headers.
Package byterange parses HTTP Range headers (RFC 9110 §14.1.2) and serves the resulting byte ranges, for handlers that cannot hand net/http.ServeContent an io.ReadSeeker — for example a handler backed by object storage, a remote origin, or computed content addressed by an Open(offset, length) call.
Package eidolon is an HTTP client for Go that reproduces the connections of real browsers: TLS ClientHello, HTTP/2 frames and settings, and header order and spelling in HTTP/2 and HTTP/1.1.
HEADERSHOT - HTTP Security Header Scanner with vulnerability explanations.
Package cloudfetch is an HTTP client that gets past Cloudflare bot detection by sending a real browser's TLS fingerprint and headers.
Package samesame implements Web Bot Auth: HTTP Message Signatures (RFC 9421) as profiled by draft-meunier-web-bot-auth-architecture, plus the key directory and Signature-Agent header from draft-meunier-http-message-signatures-directory.
Package api carries caller identity between HTTP headers and context.Context.
Package mimicry is an HTTP client that looks like a real browser on the network: the same TLS handshake, HTTP/2 and HTTP/3 settings, and headers in the same order.
Package authgate 是一个 Caddy HTTP 处理器插件,提供邮箱验证码登录、 Bearer key 认证、用户管理以及向上游注入身份 header 的能力。
Package pipeline implements an HTTP request/response middleware pipeline whose policy objects mutate an HTTP request's URL, query parameters, and/or headers before the request is sent over the wire.
Package contenttype implements HTTP Content-Type and Accept header parsers.
Package websocket implements the WebSocket protocol defined in RFC 6455. The Conn type represents a WebSocket connection. A server application calls the Upgrader.Upgrade method from an HTTP request handler to get a *Conn: Call the connection's WriteMessage and ReadMessage methods to send and receive messages as a slice of bytes. This snippet of code shows how to echo messages using these methods: In above snippet of code, p is a []byte and messageType is an int with value websocket.BinaryMessage or websocket.TextMessage. An application can also send and receive messages using the io.WriteCloser and io.Reader interfaces. To send a message, call the connection NextWriter method to get an io.WriteCloser, write the message to the writer and close the writer when done. To receive a message, call the connection NextReader method to get an io.Reader and read until io.EOF is returned. This snippet shows how to echo messages using the NextWriter and NextReader methods: The WebSocket protocol distinguishes between text and binary data messages. Text messages are interpreted as UTF-8 encoded text. The interpretation of binary messages is left to the application. This package uses the TextMessage and BinaryMessage integer constants to identify the two data message types. The ReadMessage and NextReader methods return the type of the received message. The messageType argument to the WriteMessage and NextWriter methods specifies the type of a sent message. It is the application's responsibility to ensure that text messages are valid UTF-8 encoded text. The WebSocket protocol defines three types of control messages: close, ping and pong. Call the connection WriteControl, WriteMessage or NextWriter methods to send a control message to the peer. Connections handle received close messages by calling the handler function set with the SetCloseHandler method and by returning a *CloseError from the NextReader, ReadMessage or the message Read method. The default close handler sends a close message to the peer. Connections handle received ping messages by calling the handler function set with the SetPingHandler method. The default ping handler sends a pong message to the peer. Connections handle received pong messages by calling the handler function set with the SetPongHandler method. The default pong handler does nothing. If an application sends ping messages, then the application should set a pong handler to receive the corresponding pong. The control message handler functions are called from the NextReader, ReadMessage and message reader Read methods. The default close and ping handlers can block these methods for a short time when the handler writes to the connection. The application must read the connection to process close, ping and pong messages sent from the peer. If the application is not otherwise interested in messages from the peer, then the application should start a goroutine to read and discard messages from the peer. A simple example is: Connections support one concurrent reader and one concurrent writer. Applications are responsible for ensuring that no more than one goroutine calls the write methods (NextWriter, SetWriteDeadline, WriteMessage, WriteJSON, EnableWriteCompression, SetCompressionLevel) concurrently and that no more than one goroutine calls the read methods (NextReader, SetReadDeadline, ReadMessage, ReadJSON, SetPongHandler, SetPingHandler) concurrently. The Close and WriteControl methods can be called concurrently with all other methods. Web browsers allow Javascript applications to open a WebSocket connection to any host. It's up to the server to enforce an origin policy using the Origin request header sent by the browser. The Upgrader calls the function specified in the CheckOrigin field to check the origin. If the CheckOrigin function returns false, then the Upgrade method fails the WebSocket handshake with HTTP status 403. If the CheckOrigin field is nil, then the Upgrader uses a safe default: fail the handshake if the Origin request header is present and the Origin host is not equal to the Host request header. The deprecated package-level Upgrade function does not perform origin checking. The application is responsible for checking the Origin header before calling the Upgrade function. Connections buffer network input and output to reduce the number of system calls when reading or writing messages. Write buffers are also used for constructing WebSocket frames. See RFC 6455, Section 5 for a discussion of message framing. A WebSocket frame header is written to the network each time a write buffer is flushed to the network. Decreasing the size of the write buffer can increase the amount of framing overhead on the connection. The buffer sizes in bytes are specified by the ReadBufferSize and WriteBufferSize fields in the Dialer and Upgrader. The Dialer uses a default size of 4096 when a buffer size field is set to zero. The Upgrader reuses buffers created by the HTTP server when a buffer size field is set to zero. The HTTP server buffers have a size of 4096 at the time of this writing. The buffer sizes do not limit the size of a message that can be read or written by a connection. Buffers are held for the lifetime of the connection by default. If the Dialer or Upgrader WriteBufferPool field is set, then a connection holds the write buffer only when writing a message. Applications should tune the buffer sizes to balance memory use and performance. Increasing the buffer size uses more memory, but can reduce the number of system calls to read or write the network. In the case of writing, increasing the buffer size can reduce the number of frame headers written to the network. Some guidelines for setting buffer parameters are: Limit the buffer sizes to the maximum expected message size. Buffers larger than the largest message do not provide any benefit. Depending on the distribution of message sizes, setting the buffer size to a value less than the maximum expected message size can greatly reduce memory use with a small impact on performance. Here's an example: If 99% of the messages are smaller than 256 bytes and the maximum message size is 512 bytes, then a buffer size of 256 bytes will result in 1.01 more system calls than a buffer size of 512 bytes. The memory savings is 50%. A write buffer pool is useful when the application has a modest number writes over a large number of connections. when buffers are pooled, a larger buffer size has a reduced impact on total memory use and has the benefit of reducing system calls and frame overhead. Per message compression extensions (RFC 7692) are experimentally supported by this package in a limited capacity. Setting the EnableCompression option to true in Dialer or Upgrader will attempt to negotiate per message deflate support. If compression was successfully negotiated with the connection's peer, any message received in compressed form will be automatically decompressed. All Read methods will return uncompressed bytes. Per message compression of messages written to a connection can be enabled or disabled by calling the corresponding Conn method: Currently this package does not support compression with "context takeover". This means that messages must be compressed and decompressed in isolation, without retaining sliding window or dictionary state across messages. For more details refer to RFC 7692. Use of compression is experimental and may result in decreased performance.
Package cgi implements the common gateway interface (CGI) for Caddy, a modern, full-featured, easy-to-use web server. This plugin lets you generate dynamic content on your website by means of command line scripts. To collect information about the inbound HTTP request, your script examines certain environment variables such as PATH_INFO and QUERY_STRING. Then, to return a dynamically generated web page to the client, your script simply writes content to standard output. In the case of POST requests, your script reads additional inbound content from standard input. The advantage of CGI is that you do not need to fuss with server startup and persistence, long term memory management, sockets, and crash recovery. Your script is called when a request matches one of the patterns that you specify in your Caddyfile. As soon as your script completes its response, it terminates. This simplicity makes CGI a perfect complement to the straightforward operation and configuration of Caddy. The benefits of Caddy, including HTTPS by default, basic access authentication, and lots of middleware options extend easily to your CGI scripts. CGI has some disadvantages. For one, Caddy needs to start a new process for each request. This can adversely impact performance and, if resources are shared between CGI applications, may require the use of some interprocess synchronization mechanism such as a file lock. Your server’s responsiveness could in some circumstances be affected, such as when your web server is hit with very high demand, when your script’s dependencies require a long startup, or when concurrently running scripts take a long time to respond. However, in many cases, such as using a pre-compiled CGI application like fossil or a Lua script, the impact will generally be insignificant. Another restriction of CGI is that scripts will be run with the same permissions as Caddy itself. This can sometimes be less than ideal, for example when your script needs to read or write files associated with a different owner. Serving dynamic content exposes your server to more potential threats than serving static pages. There are a number of considerations of which you should be aware when using CGI applications. CGI SCRIPTS SHOULD BE LOCATED OUTSIDE OF CADDY’S DOCUMENT ROOT. Otherwise, an inadvertent misconfiguration could result in Caddy delivering the script as an ordinary static resource. At best, this could merely confuse the site visitor. At worst, it could expose sensitive internal information that should not leave the server. MISTRUST THE CONTENTS OF PATH_INFO, QUERY_STRING AND STANDARD INPUT. Most of the environment variables available to your CGI program are inherently safe because they originate with Caddy and cannot be modified by external users. This is not the case with PATH_INFO, QUERY_STRING and, in the case of POST actions, the contents of standard input. Be sure to validate and sanitize all inbound content. If you use a CGI library or framework to process your scripts, make sure you understand its limitations. An error in a CGI application is generally handled within the application itself and reported in the headers it returns. Additionally, if the Caddy errors directive is enabled, any content the application writes to its standard error stream will be written to the error log. This can be useful to diagnose problems with the execution of the CGI application. Your CGI application can be executed directly or indirectly. In the direct case, the application can be a compiled native executable or it can be a shell script that contains as its first line a shebang that identifies the interpreter to which the file’s name should be passed. Caddy must have permission to execute the application. On Posix systems this will mean making sure the application’s ownership and permission bits are set appropriately; on Windows, this may involve properly setting up the filename extension association. In the indirect case, the name of the CGI script is passed to an interpreter such as lua, perl or python. The basic cgi directive lets you associate a single pattern with a particular script. The directive can be repeated any reasonable number of times. Here is the basic syntax: For example: When a request such as https://example.com/report or https://example.com/report/weekly arrives, the cgi middleware will detect the match and invoke the script named /usr/local/cgi-bin/report. The current working directory will be the same as Caddy itself. Here, it is assumed that the script is self-contained, for example a pre-compiled CGI application or a shell script. Here is an example of a standalone script, similar to one used in the cgi plugin’s test suite: The environment variables PATH_INFO and QUERY_STRING are populated and passed to the script automatically. There are a number of other standard CGI variables included that are described below. If you need to pass any special environment variables or allow any environment variables that are part of Caddy’s process to pass to your script, you will need to use the advanced directive syntax described below. The values used for the script name and its arguments are subject to placeholder replacement. In addition to the standard Caddy placeholders such as {method} and {host}, the following placeholder substitutions are made: - {.} is replaced with Caddy’s current working directory - {match} is replaced with the portion of the request that satisfies the match directive - {root} is replaced with Caddy’s specified root directory You can include glob wildcards in your matches. Basically, an asterisk represents a sequence of zero or more non-slash characters and a question mark represents a single non-slash character. These wildcards can be used multiple times in a match expression. See the documentation for path/Match in the Go standard library for more details about glob matching. Here is an example directive: In this case, the cgi middleware will match requests such as https://example.com/report/weekly.lua and https://example.com/report/report.lua/weekly but not https://example.com/report.lua. The use of the asterisk expands to any character sequence within a directory. For example, if the request is made, the following command is executed: Note that the portion of the request that follows the match is not included. That information is conveyed to the script by means of environment variables. In this example, the Lua interpreter is invoked directly from Caddy, so the Lua script does not need the shebang that would be needed in a standalone script. This method facilitates the use of CGI on the Windows platform. In order to specify custom environment variables, pass along one or more environment variables known to Caddy, or specify more than one match pattern for a given rule, you will need to use the advanced directive syntax. That looks like this: For example, With the advanced syntax, the exec subdirective must appear exactly once. The match subdirective must appear at least once. The env, pass_env, empty_env, and except subdirectives can appear any reasonable number of times. pass_all_env, dir may appear once. The dir subdirective specifies the CGI executable’s working directory. If it is not specified, Caddy’s current working directory is used. The except subdirective uses the same pattern matching logic that is used with the match subdirective except that the request must match a rule fully; no request path prefix matching is performed. Any request that matches a match pattern is then checked with the patterns in except, if any. If any matches are made with the except pattern, the request is rejected and passed along to subsequent handlers. This is a convenient way to have static file resources served properly rather than being confused as CGI applications. The empty_env subdirective is used to pass one or more empty environment variables. Some CGI scripts may expect the server to pass certain empty variables rather than leaving them unset. This subdirective allows you to deal with those situations. The values associated with environment variable keys are all subject to placeholder substitution, just as with the script name and arguments. If your CGI application runs properly at the command line but fails to run from Caddy it is possible that certain environment variables may be missing. For example, the ruby gem loader evidently requires the HOME environment variable to be set; you can do this with the subdirective pass_env HOME. Another class of problematic applications require the COMPUTERNAME variable. The pass_all_env subdirective instructs Caddy to pass each environment variable it knows about to the CGI excutable. This addresses a common frustration that is caused when an executable requires an environment variable and fails without a descriptive error message when the variable cannot be found. These applications often run fine from the command prompt but fail when invoked with CGI. The risk with this subdirective is that a lot of server information is shared with the CGI executable. Use this subdirective only with CGI applications that you trust not to leak this information. If you protect your CGI application with the Caddy JWT middleware, your program will have access to the token’s payload claims by means of environment variables. For example, the following token claims will be available with the following environment variables All values are conveyed as strings, so some conversion may be necessary in your program. No placeholder substitutions are made on these values. If you run into unexpected results with the CGI plugin, you are able to examine the environment in which your CGI application runs. To enter inspection mode, add the subdirective inspect to your CGI configuration block. This is a development option that should not be used in production. When in inspection mode, the plugin will respond to matching requests with a page that displays variables of interest. In particular, it will show the replacement value of {match} and the environment variables to which your CGI application has access. For example, consider this example CGI block: When you request a matching URL, for example, the Caddy server will deliver a text page similar to the following. The CGI application (in this case, wapptclsh) will not be called. This information can be used to diagnose problems with how a CGI application is called. To return to operation mode, remove or comment out the inspect subdirective. In this example, the Caddyfile looks like this: Note that a request for /show gets mapped to a script named /usr/local/cgi-bin/report/gen. There is no need for any element of the script name to match any element of the match pattern. The contents of /usr/local/cgi-bin/report/gen are: The purpose of this script is to show how request information gets communicated to a CGI script. Note that POST data must be read from standard input. In this particular case, posted data gets stored in the variable POST_DATA. Your script may use a different method to read POST content. Secondly, the SCRIPT_EXEC variable is not a CGI standard. It is provided by this middleware and contains the entire command line, including all arguments, with which the CGI script was executed. When a browser requests the response looks like When a client makes a POST request, such as with the following command the response looks the same except for the following lines: The fossil distributed software management tool is a native executable that supports interaction as a CGI application. In this example, /usr/bin/fossil is the executable and /home/quixote/projects.fossil is the fossil repository. To configure Caddy to serve it, use a cgi directive something like this in your Caddyfile: In your /usr/local/cgi-bin directory, make a file named projects with the following single line: The fossil documentation calls this a command file. When fossil is invoked after a request to /projects, it examines the relevant environment variables and responds as a CGI application. If you protect /projects with basic HTTP authentication, you may wish to enable the ALLOW REMOTE_USER AUTHENTICATION option when setting up fossil. This lets fossil dispense with its own authentication, assuming it has an account for the user. The agedu utility can be used to identify unused files that are taking up space on your storage media. Like fossil, it can be used in different modes including CGI. First, use it from the command line to generate an index of a directory, for example In your Caddyfile, include a directive that references the generated index: You will want to protect the /agedu resource with some sort of access control, for example HTTP Basic Authentication. This small example demonstrates how to write a CGI program in Go. The use of a bytes.Buffer makes it easy to report the content length in the CGI header. When this program is compiled and installed as /usr/local/bin/servertime, the following directive in your Caddy file will make it available: The cgit application provides an attractive and useful web interface to git repositories. Here is how to run it with Caddy. After compiling cgit, you can place the executable somewhere out of Caddy’s document root. In this example, it is located in /usr/local/cgi-bin. A sample configuration file is included in the project’s cgitrc.5.txt file. You can use it as a starting point for your configuration. The default location for this file is /etc/cgitrc but in this example the location /home/quixote/caddy/cgitrc. Note that changing the location of this file from its default will necessitate the inclusion of the environment variable CGIT_CONFIG in the Caddyfile cgi directive. When you edit the repository stanzas in this file, be sure each repo.path item refers to the .git directory within a working checkout. Here is an example stanza: Also, you will likely want to change cgit’s cache directory from its default in /var/cache (generally accessible only to root) to a location writeable by Caddy. In this example, cgitrc contains the line You may need to create the cgit subdirectory. There are some static cgit resources (namely, cgit.css, favicon.ico, and cgit.png) that will be accessed from Caddy’s document tree. For this example, these files are placed in a directory named cgit-resource. The following lines are part of the cgitrc file: Additionally, you will likely need to tweak the various file viewer filters such source-filter and about-filter based on your system. The following Caddyfile directive will allow you to access the cgit application at /cgit: Feeling reckless? You can run PHP in CGI mode. In general, FastCGI is the preferred method to run PHP if your application has many pages or a fair amount of database activity. But for small PHP programs that are seldom used, CGI can work fine. You’ll need the php-cgi interpreter for your platform. This may involve downloading the executable or downloading and then compiling the source code. For this example, assume the interpreter is installed as /usr/local/bin/php-cgi. Additionally, because of the way PHP operates in CGI mode, you will need an intermediate script. This one works in Posix environments: This script can be reused for multiple cgi directives. In this example, it is installed as /usr/local/cgi-bin/phpwrap. The argument following -c is your initialization file for PHP. In this example, it is named /home/quixote/.config/php/php-cgi.ini. Two PHP files will be used for this example. The first, /usr/local/cgi-bin/sample/min.php, looks like this: The second, /usr/local/cgi-bin/sample/action.php, follows: The following directive in your Caddyfile will make the application available at sample/min.php: This examples demonstrates printing a CGI rule
Package link implements parsing and serialization of Link header values as defined in RFC 5988.
Package policy — IP extraction from HTTP headers.
mcp-exposure lists the MCP servers registered in the AI clients on this machine (Claude Code, Claude Desktop, Cursor, VS Code, Gemini CLI) and reports how each one is exposed: unpinned packages fetched at every start, plaintext secrets in config files, plaintext remote transports, and local HTTP servers bound to all interfaces or accepting a foreign Host header.
Package responsecookies provides a Traefik middleware that returns a configurable HTTP response containing multiple Set-Cookie headers.
Package link implements parsing and serialization of Link header values as defined in RFC 5988.
Package gitcorsproxy is a small, sovereign reverse proxy that lets a browser reach git smart-HTTP remotes that do not send CORS headers.
Package middleware provides HTTP middleware for Go services: authentication (API key, HMAC signatures, mTLS client certificates), rate limiting, an IP allowlist, security headers, request logging, gzip compression and request body limiting.
Command goprivaudit audits a Go module's GOPRIVATE/GONOSUMDB configuration against its go.mod dependencies, git insteadOf rewrites, git credential helpers, URL-scoped extra HTTP headers, and netrc credentials, catching two silent misconfigurations: