structcli

Human-friendly, AI-native CLIs from Go structs

Declare your CLI contract once in Go structs. structcli turns it into flags, env vars, config-file loading, validation, organized help, and machine-readable contracts for agents.

• Less Cobra/Viper boilerplate
• Better CLIs for humans
• Better contracts for automation and LLMs
• Compiles to WASM out of the box

Stop writing plumbing. Start shipping commands.

⚡ Quick Start

Start with a plain Go struct:

package main

import (
	"fmt"

	"github.com/leodido/structcli"
	"github.com/spf13/cobra"
	"go.uber.org/zap/zapcore"
)

type Options struct {
	LogLevel zapcore.Level
	Port     int
}

func main() {
	opts := &Options{}
	cli := &cobra.Command{
		Use: "myapp",
		RunE: func(c *cobra.Command, args []string) error {
			fmt.Println(opts) // already populated

			return nil
		},
	}

	structcli.Bind(cli, opts)
	structcli.ExecuteOrExit(cli)
}

Bind creates flags and env vars from your struct and registers it for auto-unmarshal. ExecuteOrExit hydrates the struct from flags, env vars, config, and defaults before your RunE fires.

❯ go run examples/minimal/main.go --help
# Usage:
#   myapp [flags]
#
# Flags:
#   -h, --help                     help for myapp
#       --loglevel zapcore.Level    {debug,info,warn,error,dpanic,panic,fatal} (default info)
#       --port int

Add tags when you want aliases, env vars, shorthand, defaults, and descriptions:

type Options struct {
	LogLevel zapcore.Level `flag:"level" flagdescr:"Set logging level" flagenv:"true"`
	Port     int           `flagshort:"p" flagdescr:"Server port" flagenv:"true" default:"3000"`
}

❯ go run examples/simple/main.go -h
# A simple CLI example
#
# Usage:
#   myapp [flags]
#
# Flags:
#   -h, --help                  help for myapp
#       --level zapcore.Level   Set logging level {debug,info,warn,error,dpanic,panic,fatal} (default info)
#   -p, --port int              Server port (default 3000)
#
# Global Flags:
#       --jsonschema string[="true"]   output JSON Schema and exit (bare: this command, =tree: full subtree)
#       --mcp                          serve MCP over stdio

❯ MYAPP_LOGLEVEL=debug go run examples/simple/main.go
# &{debug 3000}

❯ MYAPP_LOGLEVEL=error MYAPP_PORT=9000 go run examples/simple/main.go --level dpanic
# &{dpanic 9000}

Built-in types like zapcore.Level are validated automatically too.

Out of the box, your CLI supports:

• 📝 Command-line flags (--level info, -p 8080)
• 🌍 Environment variables (MYAPP_PORT=8080)
• 💦 Options precedence (flags > env vars > config file > defaults)
• ✅ Automatic validation and type conversion
• 📚 Beautiful help output with proper grouping

Add the AI-native wiring below and it also gains machine-readable JSON Schema, structured JSON errors, semantic exit codes, and optional MCP tool-server mode for agents.

Build AI-Native CLIs

structcli does not just generate flags for humans. It can make your CLI legible to agents too.

Instead of scraping --help and guessing, an agent can discover the contract, call the command correctly, and recover from structured failures.

structcli.Setup(rootCmd,
    structcli.WithJSONSchema(),
    structcli.WithHelpTopics(helptopics.Options{ReferenceSection: true}),  // "mycli env-vars" and "mycli config-keys"
    structcli.WithFlagErrors(),  // Optional, but recommended for typed flag-parse errors
    structcli.WithMCP(),         // Optional, exposes the CLI as an MCP server over stdio
)
structcli.ExecuteOrExit(rootCmd)

With that wiring:

• --jsonschema exposes flags, defaults, required inputs, enums, and env bindings for the current command; --jsonschema=tree dumps the entire subtree in one call
• mycli env-vars and mycli config-keys list every environment variable binding and config file key across the command tree
• HandleError / ExecuteOrExit emit structured JSON errors instead of forcing callers to parse human-oriented output
• --mcp exposes the same command tree as MCP tools over stdio, with typed inputs and structured tool-call failures
• semantic exit codes tell the caller whether it should fix input, fix config, retry, or escalate to a human

The same contract spans flags, env vars, config, validation, and enum constraints.

$ mycli srv --jsonschema
{
  "properties": {
    "port": {
      "type": "integer",
      "default": 3000,
      "x-structcli-env-vars": ["MYCLI_SRV_PORT"]
    },
    "secret-key": {
      "type": "string",
      "x-structcli-env-vars": ["MYCLI_SRV_SECRET_KEY"],
      "x-structcli-env-only": true
    }
  },
  "x-structcli-config-flag": "config"
}

Use --jsonschema=tree to dump the entire command subtree in a single call: no need to invoke each subcommand separately:

$ mycli --jsonschema=tree   # JSON array of schemas for all commands
$ mycli srv --jsonschema=tree  # schemas for srv + its subcommands

No --help parsing. No guessing what failed. Just a CLI that can explain itself and fail in machine-actionable ways.

Use exitcode.Category(code) and exitcode.IsRetryable(code) to decide what to do next. See jsonschema.WithEnumInDescription() for schema customization, and pass schema options through WithJSONSchema with jsonschema.Options{SchemaOpts: ...}.

For CLIs that capture output streams during command construction, configure mcp.Options.CommandFactory so each MCP tool call builds a fresh command with the tool-call stdout and stderr writers. This keeps MCP protocol output separate from command output while preserving the existing command tree schema. If the command constructor requires stdin, the factory can wire a non-interactive reader such as strings.NewReader("").

For build-time discovery, generate.WriteAll produces SKILL.md, llms.txt, and AGENTS.md from the same struct definitions: wire it into //go:generate and the files stay in sync automatically.

Read the full AI-native guide or walk through the runnable structured error example.

⬇️ Install

go get github.com/leodido/structcli

📦 Key Features

🧩 Declarative Flags Definition

Define flags once using Go struct tags.

No more boilerplate for Flags().StringVarP, Flags().IntVar, viper.BindPFlag, etc.

Yes, you can nest structs too.

type ServerOptions struct {
	// Basic flags
	Host string `flag:"host" flagdescr:"Server host" default:"localhost"`
	Port int    `flagshort:"p" flagdescr:"Server port" flagrequired:"true" flagenv:"true"`

	// Environment variable binding
	APIKey string `flagenv:"true" flagdescr:"API authentication key"`

	// Network contracts using net families
	BindIP        net.IP     `flag:"bind-ip" flaggroup:"Network" flagdescr:"Bind interface IP" flagenv:"true"`
	BindMask      net.IPMask `flag:"bind-mask" flaggroup:"Network" flagdescr:"Bind interface mask" flagenv:"true"`
	AdvertiseCIDR net.IPNet  `flag:"advertise-cidr" flaggroup:"Network" flagdescr:"Advertised service subnet (CIDR)" flagenv:"true"`
	TrustedPeers  []net.IP   `flag:"trusted-peers" flaggroup:"Network" flagdescr:"Trusted peer IPs (comma separated)" flagenv:"true"`

	// Flag grouping for organized help
	LogLevel zapcore.Level `flag:"log-level" flaggroup:"Logging" flagdescr:"Set log level"`
	LogFile  string        `flag:"log-file" flaggroup:"Logging" flagdescr:"Log file path" flagenv:"true"`

	// Nested structs for organization
	Database DatabaseConfig `flaggroup:"Database"`

	// Enum type (registered via RegisterEnum)
	TargetEnv Environment `flag:"target-env" flagdescr:"Set the target environment" default:"dev"`
}

type DatabaseConfig struct {
	URL      string `flag:"db-url" flagdescr:"Database connection URL"`
	MaxConns int    `flagdescr:"Max database connections" default:"10" flagenv:"true"`
}

See full example for more details.

🛠️ Automatic Environment Variable Binding

Automatically generate environment variables binding them to configuration files (YAML, JSON, TOML, etc.) and flags.

From the previous options struct, you get the following env vars automatically:

• FULL_SRV_PORT
• FULL_SRV_APIKEY
• FULL_SRV_BIND_IP
• FULL_SRV_BIND_MASK
• FULL_SRV_ADVERTISE_CIDR
• FULL_SRV_TRUSTED_PEERS
• FULL_SRV_DATABASE_MAXCONNS
• FULL_SRV_LOGFILE, FULL_SRV_LOG_FILE

Every struct field with the flagenv:"true" tag gets an environment variable (two if the struct field also has the flag:"..." tag, see struct field LogFile). Use flagenv:"only" for fields that should be settable exclusively via environment variable or config file: CLI usage (--flag=value) is rejected at runtime.

The prefix of the environment variable name is the CLI name plus the command name to which those options are attached to.

Environment variables are command-scoped for command-local options. For example, if Port is attached to the srv command, FULL_SRV_PORT is used (not FULL_PORT).

⚙️ Configuration File Support

Set up configuration file discovery (flag, environment variable, and fallback paths) via Setup:

structcli.Setup(rootCmd,
    structcli.WithAppName("full"),
    structcli.WithConfig(config.Options{}),
)

Enable strict config-key validation with:

structcli.Setup(rootCmd,
    structcli.WithAppName("full"),
    structcli.WithConfig(config.Options{ValidateKeys: true}),
)

When enabled, Unmarshal fails if command-relevant config contains unknown keys.

WithAppName sets the env prefix. WithConfig registers the --config flag and defers config loading to ExecuteC's bind pipeline (before auto-unmarshal). Ordering between Setup and Bind does not matter: WithAppName retroactively patches env annotations on already-defined flags.

Individual SetupConfig, SetupDebug, etc. remain available for power users who need fine-grained control.

The line above:

• creates --config global flag
• creates FULL_CONFIG env var
• sets /etc/full/, $HOME/.full/, $PWD/.full/ as fallback paths for config.yaml

Magic, isn't it?

What's left? Tell your CLI to load the configuration file (if any).

rootC.PersistentPreRunE = func(c *cobra.Command, args []string) error {
	_, configMessage, configErr := structcli.UseConfigSimple(c)
	if configErr != nil {
		return configErr
	}
	if configMessage != "" {
		c.Println(configMessage)
	}

	return nil
}

UseConfigSimple(c) loads config into the root config scope and merges only the relevant section into c's effective scope.

🧠 Viper Model Scopes

structcli uses two different viper scopes on purpose:

• structcli.GetConfigViper(rootOrLeafCmd) -> root-scoped config source (config file data tree)
• structcli.GetViper(cmd) -> command-scoped effective values (flags/env/defaults + command-relevant config)

This separation keeps config-file loading isolated from runtime command state.

If you need imperative values in tests or application code, write to the right scope:

// 1) Effective override for one command context
structcli.GetViper(cmd).Set("timeout", 60)

// 2) Config-tree style injection (top-level + command section)
structcli.GetConfigViper(rootCmd).Set("srv", map[string]any{
  "port": 8443,
})

Global viper.Set(...) is not used by structcli.Unmarshal(...) resolution. Use GetViper/GetConfigViper instead.

📜 Configuration Is First-Class Citizen

Configuration can mirror your command hierarchy.

Settings can be global (at the top level) or specific to a command or subcommand. The most specific section always takes precedence.

# Global settings apply to all commands unless overridden by a specific section.
# `dryrun` matches the `DryRun` struct field name.
dryrun: true
verbose: 1 # A default verbosity level for all commands.

# Config for the `srv` command (`full srv`)
srv:
  # `port` matches the `Port` field name.
  port: 8433
  # Network options
  bind-ip: "10.20.0.10"
  bind-mask: "ffffff00"
  advertise-cidr: "10.20.0.0/24"
  trusted-peers: "10.20.0.11,10.20.0.12"
  # `log-level` matches the `flag:"log-level"` tag.
  log-level: "warn"
  # `logfile` matches the `LogFile` field name.
  logfile: /var/log/mysrv.log

  # Flattened keys can set options in nested structs.
  # `db-url` (from `flag:"db-url"` tag) maps to ServerOptions.Database.URL.
  db-url: "postgres://user:pass@db/prod"

  # Nested keys are also supported.
  database:
    # Struct field key style
    url: "postgres://user:pass@db/prod"
    # Alias key style (from `flag:"db-url"`)
    db-url: "postgres://user:pass@db/prod"

# Config for the `usr` command group.
usr:
  # This nested section matches the `usr add` command (`full usr add`).
  # Its settings are ONLY applied to 'usr add'.
  add:
    name: "Config User"
    email: "config.user@example.com"
    age: 42
    # Command specific override
    dry: false
# NOTE: Per the library's design, there is no other fallback other than from the top-level.
# A command like 'usr delete' would ONLY use the global keys above (if those keys/flags are attached to it),
# as an exact 'usr.delete' section is not defined.

This configuration system supports:

• Hierarchical Structure: Nest keys to match your command path (e.g., usr: { add: { ... } }).
• Strict Precedence: Only settings from the global scope and the exact command path section are merged. There is no automatic fallback to parent command sections.
• Flexible Keys: You can use struct field names and aliases (flag:"...") in both flattened and nested forms.
• Supported Forms for Nested Fields: db-url, database.url, database: { url: ... }, and database: { db-url: ... }.

✅ Built-in Validation & Transformation

Supports validation, transformation, and custom flag type definitions through simple interfaces.

Your struct must implement Options (via Attach) and can optionally implement ValidatableOptions and TransformableOptions.

type UserConfig struct {
	Email string `flag:"email" flagdescr:"User email" validate:"email"`
	Age   int    `flag:"age" flagdescr:"User age" validate:"min=18,max=120"`
	Name  string `flag:"name" flagdescr:"User name" mod:"trim,title"`
}

func (o *ServerOptions) Validate(ctx context.Context) []error {
    // Automatic validation
}

func (o *ServerOptions) Transform(ctx context.Context) error {
    // Automatic transformation
}

See a full working example here.

🚧 Automatic Debugging Support

Create a --debug-options flag (plus a matching env var) for troubleshooting config/env/flags resolution.

structcli.Setup(rootCmd, structcli.WithDebug(debug.Options{}))

Or standalone: structcli.SetupDebug(rootCmd, debug.Options{}).

The flag accepts text (default when used bare) or json for machine-readable output. Truthy values like true, 1, yes are treated as text for backward compatibility.

Text output: an aligned table showing each flag's resolved value and where it came from:

❯ go run examples/full/main.go srv --debug-options --config examples/full/config.yaml -p 3333
# ...
# Command: full srv
#
# Flags:
#   --apikey                 secret-api-key                       (default)
#   --config                 examples/full/config.yaml            (flag)
#   --database.maxconns      3                                    (default)
#   --db-url                 postgres://user:pass@localhost/mydb  (default)
#   --debug-options          text                                 (flag)
#   --host                   production-server                    (default)
#   --log-file               /var/log/mysrv.log                   (default)
#   --log-level              debug                                (default)
#   --port                   3333                                 (flag)
#   --target-env             dev                                  (default)
#   ...
#
# Values:
#   apikey: secret-api-key
#   host: production-server
#   log-level: debug
#   port: 3333
#   ...

JSON output: structured data for AI agents and tooling:

❯ go run examples/full/main.go srv --debug-options=json --config examples/full/config.yaml -p 3333
# ...
# {
#   "command": "full srv",
#   "flags": [
#     ...
#     {"name": "config", "value": "examples/full/config.yaml", "default": "", "changed": true, "source": "flag"},
#     {"name": "db-url", "value": "postgres://user:pass@localhost/mydb", "default": "", "changed": false, "source": "default"},
#     {"name": "log-level", "value": "debug", "default": "info", "changed": false, "source": "default"},
#     {"name": "port", "value": "3333", "default": "0", "changed": true, "source": "flag"},
#     ...
#   ],
#   "values": {"apikey": "secret-api-key", "host": "production-server", "log-level": "debug", "port": 3333, ...}
# }

Source attribution resolves each flag to flag (CLI), env, config, or default. For env-sourced flags, the text output includes the variable name (e.g., (env: MYAPP_LOG_LEVEL)).

The flag can also be activated via environment variable: FULL_DEBUG_OPTIONS=json.

📋 Self-Documenting Help Topics

WithHelpTopics (or standalone SetupHelpTopics) adds two help topic commands that list every environment variable binding and every valid configuration file key across the command tree.

structcli.Setup(rootCmd, structcli.WithHelpTopics(helptopics.Options{}))

Call this after all subcommands and flags are defined (typically right before ExecuteOrExit).

By default the commands appear as regular subcommands under "Available Commands:". Set ReferenceSection: true to move them into a dedicated "Reference:" section instead.

Environment variable listing: mycli env-vars:

Environment Variables

  mycli (global):
    MYCLI_VERBOSE  --verbose  bool           false

  mycli serve:
    MYCLI_SERVE_HOST  --host  string         localhost
    MYCLI_SERVE_PORT  --port  int            8080

Configuration key listing: mycli config-keys:

Configuration Keys

  Config flag: --config
  Supported formats: yaml, json, toml. Searches: $HOME/.mycli, /etc/mycli

  mycli (global):
    output   --output   string         text
    verbose  --verbose  bool           false

  mycli serve:
    host      --host      string         localhost
    port      --port      int            8080
    tls-cert  --tls-cert  string         ""

  Keys can be nested under the command name in the config file.

With ReferenceSection: true, both topics appear under "Reference:" in --help output instead of "Available Commands:". Flags marked flagenv:"only" show an (env-only) suffix in the env-vars listing and are excluded from config-keys (since they are hidden). Config keys derived from embedded struct field paths appear as aliases (e.g., database.maxconns → alias for --database.maxconns).

For machine-readable cross-tree data, use --jsonschema=tree instead: it provides the same information in structured JSON.

↪️ Sharing Options Between Commands

In complex CLIs, multiple commands often need access to the same global configuration and shared resources (like a logger or a database connection). structcli provides a pattern using the ContextInjector interface to achieve this without resorting to global variables, by propagating a single "source of truth" through the command context.

The pattern allows you to:

• Populate a shared options struct once from flags, environment variables, or a config file.
• Initialize "computed state" (like a logger) based on those options.
• Share this single, fully-prepared "source of truth" with any subcommand that needs it.

🍩 In a Nutshell

Create a shared struct that implements the ContextInjector interface. This struct will hold both the configuration flags and the computed state (e.g., the logger).

// This struct holds our shared state.
type CommonOptions struct {
    LogLevel zapcore.Level `flag:"loglevel" flagdescr:"Logging level" default:"info"`
    Logger   *zap.Logger   `flagignore:"true"` // This field is computed, not a flag.
}

// Context injects the struct into the command context during auto-unmarshal.
func (o *CommonOptions) Context(ctx context.Context) context.Context { /* ... */ }

// FromContext retrieves the struct from context (user-side, not interface-enforced).
func (o *CommonOptions) FromContext(ctx context.Context) error { /* ... */ }

// Initialize is a custom method to create the computed state.
func (o *CommonOptions) Initialize() error { /* ... */ }

Bind the shared struct to the root command. Bind registers it for auto-unmarshal: the bind pipeline populates it and calls Context() to inject it before any PreRunE or RunE fires.

structcli.Bind(rootC, commonOpts)

Important: Bind on root creates local flags on the root command. By default, Cobra rejects unknown flags before finding the subcommand, so app --loglevel info sub would fail. Set rootC.TraverseChildren = true so root parses its own flags first, then resolves the subcommand. Alternatively, bind the shared struct on each leaf command that needs the flags.

If you need to initialize computed state (like a logger) after unmarshal, use a PersistentPreRunE hook: by the time it fires, commonOpts is already populated:

rootC.PersistentPreRunE = func(c *cobra.Command, args []string) error {
	// commonOpts is already populated by the bind pipeline.
	return commonOpts.Initialize()
}

Retrieve the state in subcommands. In your subcommand's RunE, call .FromContext() to retrieve the shared, initialized object.

func(c *cobra.Command, args []string) error {
    config := &CommonOptions{}
    if err := config.FromContext(c.Context()); err != nil {
        return err
    }
    config.Logger.Info("Executing subcommand...")

    return nil
},

This pattern ensures that subcommands remain decoupled while having access to a consistent, centrally-managed state.

Note: The deprecated ContextOptions interface (which embeds Options and requires Attach) still works for backward compatibility. New code should use ContextInjector instead.

For a complete, runnable implementation of this pattern, see the loginsvc example located in the /examples/loginsvc directory.

🎯 Enum Registration

Register string or integer enum types once in init() and use them as plain struct fields. structcli handles flag creation, help text with allowed values, shell completion, validation, and config/env decoding automatically.

String enums (RegisterEnum)

type Environment string

const (
	EnvDev  Environment = "dev"
	EnvProd Environment = "prod"
)

func init() {
	structcli.RegisterEnum[Environment](map[Environment][]string{
		EnvDev:  {"dev", "development"},   // first string is canonical, rest are aliases
		EnvProd: {"prod", "production"},
	})
}

type DeployOptions struct {
	TargetEnv Environment `flag:"target-env" flagdescr:"Target environment" default:"dev" flagenv:"true"`
}

This produces --target-env with help text showing {dev,prod}, shell completion for all values including aliases, and case-insensitive parsing that accepts both prod and production.

Integer enums (RegisterIntEnum)

type Priority int

const (
	PriorityLow    Priority = 0
	PriorityMedium Priority = 1
	PriorityHigh   Priority = 2
)

func init() {
	structcli.RegisterIntEnum[Priority](map[Priority][]string{
		PriorityLow:    {"low"},
		PriorityMedium: {"medium", "med"},
		PriorityHigh:   {"high", "hi"},
	})
}

Both functions panic on duplicate registration or empty values. Call them in init() before any Define() calls.

See full example for enum registration in a complete CLI.

🪃 Custom Type Handlers

For types that need custom parsing logic beyond what enum registration provides, structcli offers two mechanisms: per-type registration and per-field hooks.

Per-type: RegisterType[T]

Register custom hooks for a type once in init(). Every struct field of type T uses them automatically: no special tags needed. RegisterType panics on nil hooks or duplicate registration: call it in init() before any Define/Bind.

import (
    "github.com/leodido/structcli"
    "github.com/leodido/structcli/values"
)

type ListenAddress struct {
    Host string
    Port int
    raw  string
}

func init() {
    structcli.RegisterType[ListenAddress](structcli.TypeHooks[ListenAddress]{
        Define: func(name, descr string, sf reflect.StructField, fv reflect.Value) (pflag.Value, string) {
            ptr := fv.Addr().Interface().(*ListenAddress)
            *ptr = ListenAddress{Host: "localhost", Port: 8080, raw: "localhost:8080"}

            return values.NewString(&ptr.raw), descr + " (host:port)"
        },
        Decode: func(input any) (any, error) {
            return ParseListenAddress(input.(string))
        },
    })
}

type ServerOptions struct {
    ListenAddr ListenAddress `flag:"listen" flagdescr:"Listen address"`
}

Both Define and Decode are required. Panics on duplicate registration or nil hooks. Call in init() before any Define/Bind calls.

For enum types, prefer RegisterEnum/RegisterIntEnum: they wrap RegisterType with less boilerplate.

Per-field: FieldHookProvider and FieldCompleter

When the same type needs different behavior in different fields, or when a standard type needs custom handling for a specific field, implement FieldHookProvider on your options struct. Map keys are struct field names.

type ServerOptions struct {
    ListenAddr string `flag:"listen" flagdescr:"Listen address"`
    Mode       string `flag:"mode" flagdescr:"Server mode"`
}

func (o *ServerOptions) FieldHooks() map[string]structcli.FieldHook {
    return map[string]structcli.FieldHook{
        "ListenAddr": {
            Define: func(name, descr string, sf reflect.StructField, fv reflect.Value) (pflag.Value, string) {
                ptr := fv.Addr().Interface().(*string)
                *ptr = "localhost:8080"

                return values.NewString(ptr), descr + " (host:port)"
            },
            Decode: func(input any) (any, error) {
                return input, nil
            },
        },
    }
}

For shell completion, implement FieldCompleter:

func (o *ServerOptions) CompletionHooks() map[string]structcli.CompleteHookFunc {
    return map[string]structcli.CompleteHookFunc{
        "ListenAddr": func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
            return []string{"localhost:8080", "0.0.0.0:8080", "0.0.0.0:443"}, cobra.ShellCompDirectiveNoFileComp
        },
    }
}

FieldCompleter works for any field that becomes a flag, not only fields with FieldHookProvider hooks.

Precedence

Hook resolution follows this order (highest to lowest):

1. FieldHookProvider: per-field hooks on the options struct
2. RegisterType / RegisterEnum: per-type hooks in the global registry
3. Built-in registry: time.Duration, zapcore.Level, slog.Level, etc.

Completion precedence:

• If a completion function is already registered on a flag before Define, structcli preserves it.
• If Define auto-registers a FieldCompleter hook, a later manual RegisterFlagCompletionFunc on the same flag returns Cobra's already registered error.

In values we provide pflag.Value implementations for standard types.

See the custom types example for a runnable demo of all three mechanisms, or the full example for a complete CLI.

🧱 Built-in Custom Types

Type	Description	Example Values	Special Features
zapcore.Level	Zap logging levels	debug, info, warn, error, dpanic, panic, fatal	Enum validation
slog.Level	Standard library logging levels	debug, info, warn, error, error+2, ...	Level offsets: ERROR+2, INFO-4
time.Duration	Time durations	30s, 5m, 2h, 1h30m	Go duration parsing
[]time.Duration	Duration slices	30s,5m, 1s,2m30s	Comma-separated / repeated flags
[]bool	Boolean slices	true,false,true	Comma-separated / repeated flags
[]uint	Unsigned integer slices	1,2,3,42	Comma-separated / repeated flags
[]byte	Raw textual bytes	hello, abc123	Raw textual input
structcli.Hex	Hex-decoded textual input	68656c6c6f, 48656c6c6f	Hex decoding
structcli.Base64	Base64-decoded textual input	aGVsbG8=, YWJjMTIz	Base64 decoding
net.IP	IP address	127.0.0.1, 10.42.0.10, 2001:db8::1	IP parsing
net.IPMask	IPv4 mask	255.255.255.0, ffffff00	Dotted or hex mask parsing
net.IPNet	CIDR subnet	10.42.0.0/24, 2001:db8::/64	CIDR parsing
[]net.IP	IP slices	10.0.0.1,10.0.0.2	Comma-separated / repeated flags
[]string	String slices	item1,item2,item3	Comma-separated
[]int	Integer slices	1,2,3,42	Comma-separated
map[string]string	String maps	env=prod,team=platform	key=value pairs
map[string]int	Integer maps	cpu=2,memory=4	key=value pairs with int parsing
map[string]int64	64-bit integer maps	ok=1,fail=2	key=value pairs with int64 parsing

Note on JSON output: net.IPMask is a byte slice under the hood, so Go's encoding/json renders it as base64 (for example 255.255.255.0 appears as ////AA==). This is expected.

All built-in types support:

• Command-line flags with validation and help text
• Environment variables with automatic binding
• Configuration files (YAML, JSON, TOML)
• Type validation with helpful error messages

Slices and maps use the same contract across flags, env vars, and config.

See examples/collections/main.go for a runnable version of this example.

type AdvancedOptions struct {
	Retries   []uint          `flag:"retries" flagenv:"true"`
	Backoffs  []time.Duration `flag:"backoffs" flagenv:"true"`
	FeatureOn []bool          `flag:"feature-on" flagenv:"true"`
	Labels    map[string]string `flag:"labels" flagenv:"true"`
	Limits    map[string]int    `flag:"limits" flagenv:"true"`
	Counts    map[string]int64  `flag:"counts" flagenv:"true"`
}

❯ myapp --retries 1,2,3 --backoffs 1s,5s --feature-on true,false --labels env=prod,team=platform --limits cpu=8,memory=16 --counts ok=10,fail=3
❯ MYAPP_RETRIES=1,2,3 MYAPP_BACKOFFS=1s,5s MYAPP_FEATURE_ON=true,false MYAPP_LABELS=env=prod,team=platform MYAPP_LIMITS=cpu=8,memory=16 MYAPP_COUNTS=ok=10,fail=3 myapp
❯ go run examples/collections/main.go --config examples/collections/config.yaml

retries: "1,2,3"
backoffs:
  - 1s
  - 5s
feature-on: "true,false"
labels:
  env: prod
  team: platform
limits:
  cpu: 8
  memory: 16
counts: "ok=10,fail=3"

🧰 Reusable Flag Kits

The flagkit package provides pre-built, embeddable flag structs that standardize common CLI flag declarations. Each type encapsulates one flag with an opinionated name, type, and default matching industry conventions. This gives AI agents and scripts a consistent vocabulary across CLIs built with structcli.

import "github.com/leodido/structcli/flagkit"

type LogsOptions struct {
    flagkit.Follow                                                    // --follow/-f (default: false)
    Service string `flag:"service" flagshort:"s" flagdescr:"Service name" flagrequired:"true"`
}

func (o *LogsOptions) Attach(c *cobra.Command) error {
    if err := structcli.Define(c, o); err != nil {
        return err
    }
    flagkit.AnnotateCommand(c) // marks flagkit-owned flags for doc generation
    return nil
}

Attach is needed here because of the custom flagkit.AnnotateCommand call. For structs without custom define-time logic, use structcli.Bind(cmd, opts) instead: it handles both plain structs and Options implementors.

Available types:

Type	Flag	Default	Description
Follow	--follow / -f	false	Opt-in streaming (agents won't hang)
LogLevel	--log-level	info	Log level via zapcore (alias for ZapLogLevel)
ZapLogLevel	--log-level	info	Log level backed by zapcore.Level
SlogLogLevel	--log-level	info	Log level backed by slog.Level (stdlib)
Output	--output / -o	text	Output format (string enum, user-registered)
Verbose	--verbose / -v	0	Verbosity count (-v, -vv, -vvv)
DryRun	--dry-run	false	Preview without making changes
Timeout	--timeout	30s	Operation timeout (time.Duration)
Quiet	--quiet / -q	false	Suppress non-essential output

When the generate package detects flagkit annotations, it emits a "Development Notes" section in AGENTS.md guiding AI coding agents to prefer flagkit types over ad-hoc flag declarations.

See go doc github.com/leodido/structcli/flagkit for the full taxonomy and composition examples.

🎨 Beautiful, Organized Help Output

Organize your --help output into logical groups for better readability.

❯ go run examples/full/main.go --help
# A demonstration of the structcli library with beautiful CLI features
#
# Usage:
#   full [flags]
#   full [command]
#
# Available Commands:
#   completion  Generate the autocompletion script for the specified shell
#   help        Help about any command
#   logs        Show service logs
#   preset      Demonstrate flag presets with validation and transformation
#   srv         Start the server
#   usr         User management
#
# Flags:
#   -h, --help   help for full
#
# Utility Flags:
#       --dry             
#   -v, --verbose count
#
# Global Flags:
#       --config string                   config file (fallbacks to: {/etc/full,{executable_dir}/.full,$HOME/.full,...}/config.{yaml,json,toml})
#       --debug-options string[="text"]   debug output format (text, json)
#       --jsonschema string[="true"]      output JSON Schema and exit (bare: this command, =tree: full subtree)
#       --mcp                             serve MCP over stdio
#
# Reference:
#   config-keys List all configuration file keys
#   env-vars    List all environment variable bindings

❯ go run examples/full/main.go srv --help
# Start the server with the specified configuration
#
# Usage:
#   full srv [flags]
#   full srv [command]
#
# Available Commands:
#   version     Print version information
#
# Flags:
#       --apikey string                  API authentication key
#       --deep-setting string             (default "default-deep-setting")
#       --deep.deeper.nodefault string
#       --deeper-setting string           (default "default-deeper-setting")
#   -h, --help                           help for srv
#       --host string                    Server host (default "localhost")
#   -p, --port int                       Server port
#       --target-env string              Set the target environment {dev,prod,staging} (default "dev")
#
# Database Flags:
#       --database.maxconns int   Max database connections (default 10)
#       --db-url string           Database connection URL
#
# Logging Flags:
#       --log-file string           Log file path
#       --log-level zapcore.Level   Set log level {debug,info,warn,error,dpanic,panic,fatal} (default info)
#
# Network Flags:
#       --advertise-cidr ipNet    Advertised service subnet (CIDR) (default 127.0.0.0/24)
#       --bind-ip ip              Bind interface IP (default 127.0.0.1)
#       --bind-mask ipMask        Bind interface mask (default ffffff00)
#       --trusted-peers ipSlice   Trusted peer IPs (comma separated) (default 127.0.0.2,127.0.0.3)
#
# Security Flags:
#       --token-base64 bytesBase64   Token bytes encoded as base64 (default aGVsbG8=)
#       --token-hex bytesHex         Token bytes encoded as hex (default 68656c6c6f)
#
# Global Flags:
#       --config string                   config file (fallbacks to: {/etc/full,{executable_dir}/.full,$HOME/.full,...}/config.{yaml,json,toml})
#       --debug-options string[="text"]   debug output format (text, json)
#       --jsonschema string[="true"]      output JSON Schema and exit (bare: this command, =tree: full subtree)
#       --mcp                             serve MCP over stdio
#
# Use "full srv [command] --help" for more information about a command.

🌐 WebAssembly (WASM)

structcli avoids reflect.Value.MethodByName and reflect.Value.Call, so the Go compiler's dead-code elimination (DCE) works correctly and WASM binaries stay lean. All features work under GOOS=wasip1 GOARCH=wasm with the standard Go compiler, including custom type hooks (RegisterType[T], FieldHookProvider), JSON Schema generation, structured errors, and debug output.

Build and run any structcli program as WASM:

GOOS=wasip1 GOARCH=wasm go build -o myapp.wasm .
wasmtime run --dir=/ myapp.wasm serve --port 8080

This means structcli CLIs can run in sandboxed WASM runtimes, edge functions, and browser-based environments. Combined with --jsonschema and --mcp, an agent can discover the CLI contract, call it in a sandboxed WASM runtime, and handle structured failures without ever touching the host OS.

🏷️ Available Struct Tags

Use these tags in your struct fields to control the behavior:

Tag	Description	Example
flag	Sets a custom name for the flag (otherwise, generated from the field name)	flag:"log-level"
flagpreset	Defines CLI-only preset aliases for this field's flag. Each preset is <alias-flag-name>=<value-for-this-field-flag>. No env/config keys are created.	flagpreset:"logeverything=5;logquiet=0"
flagshort	Sets a single-character shorthand for the flag	flagshort:"l"
flagdescr	Provides the help text for the flag	flagdescr:"Logging level"
default	Sets the default value for the flag	default:"info"
flagenv	Enables binding to an environment variable ("true", "false", or "only")	flagenv:"true"
flagrequired	Marks the flag as required ("true"/"false")	flagrequired:"true"
flaghidden	Hides the flag from help/usage output and machine-readable schemas while keeping it fully functional ("true"/"false")	flaghidden:"true"
flaggroup	Assigns the flag to a group in the help message	flaggroup:"Database"
flagignore	Skips creating a flag for this field ("true"/"false")	flagignore:"true"
flagtype	Specifies a special flag type. Currently supports count	flagtype:"count"

flagpreset is syntactic sugar: it creates alias flags that set the canonical flag value. Format: <alias>=<value>; multiple entries can be separated by ; or ,. Example: flagpreset:"logeverything=5;logquiet=0" makes --logeverything behave like --loglevel=5. If both alias and canonical flags are passed, the last assignment in argv wins. It does not bypass transform/validate flow.

flaghidden:"true" + flagenv:"true" vs flagenv:"only":

• flaghidden:"true" + flagenv:"true": hidden from help, but accepts CLI input via --flag=value. Use for flags that should be discoverable only by advanced users or scripts.
• flagenv:"only": hidden from help, rejects CLI input at runtime. The field is settable only via environment variable or config file. Use for secrets and deployment-time configuration that should never appear on a command line.

flagenv:"only" is incompatible with flagshort, flagpreset, and flagtype (these are CLI-only concepts). It supports flagdescr, flaggroup, flagrequired, and default. FieldHookProvider Define/Decode hooks work normally on flagenv:"only" fields (the flag is created then hidden). FieldCompleter hooks are skipped since hidden flags have no CLI completion.

📖 Documentation

For comprehensive documentation and advanced usage patterns, visit the documentation.

Start here for repo-local guides:

• AI-Native CLIs guide
• Structured error example walkthrough
• Examples directory

🤝 Contributing

Contributions are welcome!

Please feel free to submit a Pull Request.