Cursor Agent CLI support

July 23, 2026 · View on GitHub

Minutes supports the Cursor Agent CLI (agent) as an opt-in local agent for non-interactive summarization.

What is wired

  • Desktop settings recognize agent as a well-known agent_command when the binary is on PATH (or in well-known install dirs such as ~/.local/bin).
  • engine = "agent" can run agent for post-meeting summarization, title refinement, and L1 speaker mapping.
  • Cursor continues to work as a files + MCP host for interactive chat; this doc covers only the headless agent_command path.
  • No separate Cursor skill tree — reuse raw ~/meetings/ files and/or the Minutes MCP server / portable .agents/skills mirror.

Summarization config

[summarization]
engine = "agent"
agent_command = "agent"

Use an absolute path when the desktop app PATH is minimal:

agent_command = "/Users/you/.local/bin/agent"

Minutes runs Cursor Agent with:

agent -p --mode ask --output-format text --trust "<prompt>"

That invocation is intentionally narrow:

  • -p / print mode for non-interactive stdout capture
  • --mode ask for read-only Q&A (no write or shell tools as a side effect)
  • --trust so headless runs do not block on an interactive workspace-trust prompt (same class of bypass as Codex --skip-git-repo-check / Gemini --skip-trust)
  • prompt as a trailing positional argument (CLI contract)
  • no --force / --yolo
  • no hardcoded --model — model selection stays with your Cursor CLI / account (Pro/Ultra subscription or CURSOR_API_KEY auth as configured for the CLI)

Authenticate the CLI first (agent login, or set CURSOR_API_KEY for print mode). Interactive IDE login alone may not be enough for headless -p runs.

Screens / live coaching

Headless Cursor Agent summarization is text-only in this release (same posture as gemini/opencode/pi). Screen-context delivery for agent is not wired yet.