Skip to content

types

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

types

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
README.md
README.md
 
 
doc.go
doc.go
 
 
mcp.go
mcp.go
 
 
spec_test.go
spec_test.go
 
 
token_weights.go
token_weights.go
 
 

README.md

types Package

The types package provides shared type definitions used across multiple gh-aw packages to avoid circular dependencies.

Overview

This package defines common data structures that are shared between the parser and workflow packages. Centralizing these types here allows both packages to reference the same definitions without creating import cycles.

Types

BaseMCPServerConfig

The foundational configuration structure for MCP (Model Context Protocol) servers. This type is embedded by both parser.MCPServerConfig and workflow.MCPServerConfig.

MCP servers can run as:

  • stdio processes: Command + Args, launched as a child process.
  • HTTP endpoints: URL + optional Headers and Auth, reached over HTTP/HTTPS.
  • Container services: Container image + optional Mounts, run inside a container.
import "github.com/github/gh-aw/pkg/types"

// Stdio MCP server
cfg := types.BaseMCPServerConfig{
    Type:    "stdio",
    Command: "npx",
    Args:    []string{"-y", "@modelcontextprotocol/server-filesystem"},
    Env: map[string]string{
        "ALLOWED_PATHS": "/workspace",
    },
}

// HTTP MCP server with OIDC auth
cfg := types.BaseMCPServerConfig{
    Type: "http",
    URL:  "https://my-mcp-server.example.com",
    Auth: &types.MCPAuthConfig{
        Type:     "github-oidc",
        Audience: "https://my-mcp-server.example.com",
    },
}

Fields

Field Type Description
Type string Server type: "stdio", "http", "local", or "remote"
Command string Executable to launch (stdio mode)
Args []string Arguments passed to the command
Env map[string]string Environment variables injected into the process
Version string Optional version or tag
URL string HTTP endpoint URL (HTTP mode)
Headers map[string]string Additional HTTP headers (HTTP mode)
Auth *MCPAuthConfig Upstream authentication (HTTP mode only)
Container string Container image (container mode)
Entrypoint string Optional entrypoint override for the container
EntrypointArgs []string Arguments passed to the container entrypoint
Mounts []string Volume mounts in "source:dest:mode" format

MCPAuthConfig

Authentication configuration for HTTP MCP servers. When configured, the MCP gateway dynamically acquires tokens and injects them as Authorization headers on each outgoing request.

auth := &types.MCPAuthConfig{
    Type:     "github-oidc", // Currently the only supported type
    Audience: "https://my-service.example.com",
}
Field Type Description
Type string Auth type; currently only "github-oidc" is supported
Audience string OIDC token audience (aud claim); defaults to the server URL if omitted

TokenWeights

Defines custom model cost information for effective token computation. Specified under engine.token-weights in workflow frontmatter and stored in aw_info.json at runtime.

weights := types.TokenWeights{
    Multipliers: map[string]float64{
        "gpt-4o": 2.5,
    },
    TokenClassWeights: &types.TokenClassWeights{
        Input:  1.0,
        Output: 3.0,
    },
}

TokenClassWeights

Per-token-class weights for effective token computation. Each field corresponds to one token class; a zero value means "use the default weight".

Field Token class
Input Standard input tokens
CachedInput Cache-hit input tokens
Output Generated output tokens
Reasoning Internal reasoning tokens
CacheWrite Cache-write tokens

Design Notes

  • This package has no dependencies on other gh-aw packages, making it safe to import from anywhere.
  • All struct fields use both json and yaml struct tags so they can be round-tripped through both serialization formats.
  • BaseMCPServerConfig is designed to be embedded — packages add domain-specific fields and validation on top of the shared base.

This specification is automatically maintained by the spec-extractor workflow.