Setting up Codex through ACP

Install Node.js with npm and make sure npx is available to GT. Install Codex using the official OpenAI documentation, then run codex login in a terminal to sign in. To let the agent work with the running image, set up GtLMcpServer Object << #GtLMcpServer slots: { #port . #tools . #resources . #prompts . #server . #history . #errors . #debug . #asynchronous . #currentAsyncExecution . #objectStorage . #summarizersRegistry . #announcer }; sharedVariables: { #Current }; tag: 'Mcp'; package: 'Gt4LlmCore' following Working with GT as an MCP server and Adding tools to the GT MCP server, then connect the agent to it.

Create a GtLACPCodexClient GtLACPClient << #GtLACPCodexClient slots: {}; tag: 'Core'; package: 'Gt4Llm-ACP' with a request timeout of 180 seconds:

client := GtLACPCodexClient new.
client configuration requestTimeout: 180.
client
  

Replace the path with an existing project directory, then create and inspect a session:

directory := '/absolute/path/to/your/project' asFileReference.
session := client startSessionIn: directory.
session
  

Use the Chat view to send prompts and respond to permission requests. When finished, close the session:

session close
  

See Loading ACP sessions from disk to continue an existing conversation, or Setting up Claude Code through ACP to use Claude Code.