HTTP Request and Response Parameter Mapping (MoonBit)

Overview

When an agent is exposed over HTTP, Golem maps parts of each HTTP request to constructor and method parameters. This skill covers how path segments, query parameters, headers, and request bodies are mapped, which types are supported for each, and how return types map to HTTP responses.

Path Variables

Path variables {var_name} in mount or endpoint paths map to parameters by name:

///|
/// A task management agent
#derive.agent
#derive.mount("/api/tasks/{task_name}")
pub(all) struct TaskAgent {
  task_name : String
}
 
///|
fn TaskAgent::new(task_name : String) -> TaskAgent {
  { task_name }
}
 
///|
/// Get an item by ID
#derive.endpoint(get="/items/{item_id}")
pub fn TaskAgent::get_item(self : Self, item_id : String) -> Item {
  // task_name from mount path, item_id from endpoint path
  ...
}

Remaining (catch-all) path variables capture everything after a prefix:

#derive.endpoint(get="/files/{*path}")
pub fn TaskAgent::get_file(self : Self, path : String) -> FileContent {
  // GET .../files/docs/readme.md → path = "docs/readme.md"
  ...
}

Catch-all variables can only appear as the last path segment and are not allowed in mount paths.

Query Parameters

Specified in the endpoint path using ?key={var} syntax:

#derive.endpoint(get="/search?q={query}&limit={max_results}")
pub fn TaskAgent::search(self : Self, query : String, max_results : UInt64) -> Array[SearchResult] {
  // GET .../search?q=hello&limit=10
  ...
}

Header Variables

Map HTTP headers to parameters using #derive.endpoint_header:

#derive.endpoint(get="/data")
#derive.endpoint_header("X-Request-Id", "request_id")
#derive.endpoint_header("Authorization", "token")
pub fn TaskAgent::get_data(self : Self, request_id : String, token : String) -> Data {
  ...
}

Each #derive.endpoint_header maps one header to one method parameter. Duplicate header names (case-insensitive) are not allowed.

Supported Types for Path, Query, and Header Variables

Only these types can be used for parameters bound to path/query/header variables (the value is parsed from the URL/header string):

For query parameters and headers only (not path variables), two additional wrapper types are supported:

All other types (structs, tuples, enums with data, etc.) can only be used as body parameters.

POST Request Body Mapping

User-supplied method parameters not bound to path variables, query parameters, or headers are body parameters. Ordinary structured parameters are populated from the JSON request body:

#derive.endpoint(post="/items/{id}")
pub fn TaskAgent::update_item(self : Self, id : String, name : String, count : UInt64) -> Item {
  // POST .../items/123
  // Body: { "name": "Widget", "count": 5 }
  // → id from path, name and count from body
  ...
}

Each ordinary structured body parameter becomes a top-level field in the expected JSON body object. All custom types must have #derive.golem_schema.

⚠️ Important: Ordinary structured bodies use a JSON object with body parameter names only as keys, even for one parameter. decide(self : Self, decision : String) expects {"decision": "approved"}, not a bare string. A single object parameter named payload also needs wrapping: {"payload": {"name": "Widget", "count": 5}}, not {"name": "Widget", "count": 5}. Do not include path/query/header parameters. Unstructured text and binary parameters instead consume the raw body; see below.

CLI positional invocation is different: pass one argument per user-supplied method parameter in signature order, including parameters bound to path/query/header values over HTTP. Skip host-injected Principal fields without a principal value or placeholder. Use MoonBit native literals for each value, not one JSON object wrapping the full input. A single object parameter is one native struct literal without the HTTP payload wrapper. Exclude self and config parameters. Load golem-invoke-agent-moonbit for CLI examples and caller-only input rules.

Raw Text and Binary Bodies

Use @schema.UnstructuredText[L] for raw text/plain and @schema.UnstructuredBinary[M] for raw binary request and response bodies. Each is inline content or a URL reference:

pub(all) enum UnstructuredText[L] {
  Inline(text~ : String, language~ : String?)
  Url(String)
}
 
pub(all) enum UnstructuredBinary[M] {
  Inline(bytes~ : Bytes, mime_type~ : String?)
  Url(String)
}

Restrictions are type-level: the type argument implements @schema.AllowedLanguages (languages()) or @schema.AllowedMimeTypes (mime_types()). Use @schema.AnyLanguage / @schema.AnyMimeType for no restriction. There are no #derive.text_languages / #derive.mime_types annotations.

///|
pub(all) struct ImageTypes {}
 
///|
impl @schema.AllowedMimeTypes for ImageTypes with fn mime_types() {
  ["image/png", "image/gif"]
}
 
///|
#derive.endpoint(post="/upload/{bucket}")
pub fn UploadAgent::upload(
  self : Self,
  bucket : String,
  payload : @schema.UnstructuredBinary[ImageTypes],
) -> Int {
  match payload {
    Inline(bytes~, ..) => bytes.length()
    Url(_) => -1
  }
}
 
///|
#derive.endpoint(get="/notes/{id}")
pub fn UploadAgent::get_note(
  self : Self,
  id : String,
) -> @schema.UnstructuredText[@schema.AnyLanguage] {
  @schema.UnstructuredText::inline("hello", language="en")
}

• Construct values with UnstructuredText::inline(text, language?) / UnstructuredBinary::inline(bytes, mime_type?) or ::url(url).
• A method with an unstructured body parameter may have only one body parameter, and it cannot be bound to a path/query/header variable.
• Text request bodies must have no Content-Type, text/plain, or text/plain; charset=utf-8 (otherwise 415); an optional single Content-Language is checked against the allow-list.
• Responses: text is sent as text/plain; charset=utf-8 with the language as Content-Language; binary uses its MIME type (default application/octet-stream); a Url value is a 307 redirect.

Return Type to HTTP Response Mapping

The value inside Some / Ok follows the same rules as a top-level value: an unstructured value nested in T?, Result[T, E], or Result[T?, E] is sent as the raw body, not as JSON. For example, Result[@schema.UnstructuredBinary[M]?, String] gives raw bytes (200), 404 for Ok(None), and a JSON error (500) for Err.

Data Type to JSON Mapping

Custom Types

All custom structs and enums used as parameters or return types must be annotated with #derive.golem_schema:

#derive.golem_schema
pub(all) enum Priority {
  Low
  Medium
  High
} derive(Eq)
 
#derive.golem_schema
pub(all) struct TaskInfo {
  title : String
  priority : Priority
  description : String?
}

Complete Example

///|
/// A REST API agent exposing weather data via HTTP endpoints
#derive.agent
#derive.mount("/api/{city}/weather")
pub(all) struct WeatherAgent {
  city : String
  mut last_temperature : Double
}
 
///|
fn WeatherAgent::new(city : String) -> WeatherAgent {
  { city, last_temperature: 0.0 }
}
 
///|
/// Get current temperature
#derive.endpoint(get="/current?unit={unit}")
pub fn WeatherAgent::get_temperature(self : Self, unit : String) -> Double {
  if unit == "fahrenheit" {
    self.last_temperature * 9.0 / 5.0 + 32.0
  } else {
    self.last_temperature
  }
}
 
///|
/// Set the temperature
#derive.endpoint(post="/set")
#derive.endpoint_header("X-Source", "source")
pub fn WeatherAgent::set_temperature(
  self : Self,
  temperature : Double,
  source : String,
) -> String {
  self.last_temperature = temperature
  "Temperature set to " + temperature.to_string() + " from " + source
}

Related Guides

• Load golem-add-http-endpoint-moonbit for the high-level workflow of defining and mounting HTTP endpoints on an agent