Expose MAX mode for headless cursor-agent --print execution

Feature request for product/service

Cursor CLI

Describe the request

Cursor CLI supports MAX mode interactively through /max-mode and the persisted CLI setting, but MAX cannot currently be selected for headless cursor-agent --print executions.

This makes it impossible for external tooling and automation to request MAX mode reliably on a per-run basis. Disabling --print and automating an interactive terminal is substantially less reliable because completion, cancellation, permissions, output parsing, and resumed-session state become harder to control.

Requested solution

Please expose MAX mode through a supported headless interface, for example:

cursor-agent --print --max-mode ...

or through a structured per-request configuration option.

Desired behavior:

  • usable together with --print;
  • selectable per invocation;
  • reflected in structured output or metadata;
  • explicit error when the selected model does not support MAX;
  • deterministic behavior for new and resumed conversations;
  • no need to mutate the global ~/.cursor/cli-config.json;
  • capability visible through --help or a machine-readable capabilities command.

Use case

I am building a local bridge that invokes Cursor CLI as an execution agent. The bridge supports model and execution-mode selection, approval gates, cancellation, permissions, and conversation resume. Normal headless execution works well through --print, but MAX cannot be exposed safely because the CLI only offers it interactively.

An official headless option would allow integrations to support MAX without fragile pseudo-terminal automation.

Hey @Ivar_Koster!

Max Mode is going away for users on usage-based plans.

If you want to use a larger context window via the CLI, you should be able to just pass that info to --model.

cursor-agent --print --model 'claude-opus-4-8[thinking=true,context=1m,effort=high,fast=true]'