App Server
Open Interpreter uses an app-server-backed runtime for the interactive TUI. The same app-server protocol is available for advanced integrations that need threads, streamed events, approvals, session state, model selection, and MCP status inside another application.
Most automation should start with Non-interactive mode or the SDK. Use the app server when you are building a richer client. Use Agent Client Protocol instead when you want an ACP-compatible editor or UI to launch Open Interpreter as its coding agent.
Start a Local Server
Use stdio for an SDK-style child process:
interpreter-app-server --listen stdio://Use WebSocket transport when a separate client needs to connect:
interpreter-app-server --listen ws://127.0.0.1:9000Then connect a TUI client:
interpreter --remote ws://127.0.0.1:9000Secure Remote Access
If the server is not strictly local, terminate TLS in front of it and require a bearer token:
interpreter --remote wss://agent.example.com \
--remote-auth-token-env INTERPRETER_REMOTE_TOKENStore the token in the named environment variable on the client. Do not expose an unauthenticated app-server listener on a public network.
Protocol
The protocol is inherited from Codex app-server. Open Interpreter keeps that
surface compatible so existing Codex app-server clients can use
interpreter-app-server as the launched process.
Use OpenAI's Codex app-server docs for the full protocol shape, then apply the Open Interpreter-specific launch commands above.