Quey is available on desktop

Approval modes

When you send a capture, the approval mode controls whether the agent runs immediately or waits for your approval. Choose based on how much control you want and how confident you are in the instruction.

The three modes

Ask

Always pause before the agent starts. You see the capture and approve before any files are touched.

  • Best for: learning, exploring, or untested instructions
  • Flow: capture → review pre-run → approve → agent runs → review diff → approve changes
  • Safety: highest — you control both pre-run and post-run
Note
Use Ask mode when you want to double-check the context the agent sees before it touches your files.

Smart

Pause only for runs that will edit files in your workspace. Safe default for mixed sessions (e.g., inspecting design tools while editing your codebase).

  • Best for: balanced workflow with most captures safe and some risky
  • Flow: capture → (if risky, pause for approval) → agent runs → review diff → approve changes
  • Safety: medium — you approve risky edits but not read-only inspections
Tip
Smart mode is the default. It skips approval for captures that won't edit files (like inspect mode).

Allow

Run immediately without pre-execution approval. Still shows the diff after, so you can review and reject if needed.

  • Best for: fast iteration when you're confident, or fully hands-free auto-run
  • Flow: capture → agent runs immediately → review diff → approve/reject changes
  • Safety: lowest — relies on your post-run review
Warning
Allow mode trusts your instruction. If you send something unclear, the agent will interpret it and make changes before asking. Always review the diff.

Choosing a mode

Set the approval mode from the Agent panel's dropdown in the toolbar. It's attached to each capture when you send it, so changing it mid-session only affects new captures — not ones already in flight.

Note
There's no environment variable to set a project-wide default — approval mode is a per-capture, client-side choice, not a runner setting. See configuration for the environment variables that are available.

Auto-run (fully hands-free)

Auto-run is on by default — the only thing missing is QUEY_CODEX_CWD, the absolute path to the repo you want the agent to edit. Without it, auto-run has nowhere to write and each capture just records an error asking you to set it:

.mcp.jsonjson
{  "mcpServers": {    "quey": {      "command": "npx",      "args": ["-y", "@quey/mcp", "server"],      "env": {        "QUEY_CODEX_CWD": "/absolute/path/to/repo"      }    }  }}

With auto-run active:

  1. You send a capture from the browser, with whatever approval mode is selected in the Agent panel.
  2. The MCP runner detects it and launches Claude Code or Codex automatically — whichever provider the capture targets.
  3. The agent edits the file and marks the capture as resolved.
  4. The diff appears in the toolbar for your review.
Note
Set QUEY_AUTO_RUN=0 to disable auto-run entirely (e.g. to drive Claude/Codex manually via MCP tools instead). Cursor doesn't use auto-run at all — it reads pending captures directly through MCP tools, so neither variable matters for it.

For a truly hands-free loop where your agent watches for captures continuously, use quey_watch in your agent's automation:

examplepython
while True:    capture = quey_watch()  # blocks until a new capture arrives    # process capture    quey_resolve(capture.id, summary="Applied changes")
Warning
Auto-run requires QUEY_CODEX_CWD (absolute path to your repo) to be set. Without it, auto-run is disabled.

When to use each mode

  • Ask: First time with a tool, testing new instructions, high-risk edits
  • Smart: Daily development, mixed sessions, most captures are safe
  • Allow: Repetitive edits, hands-free workflows, trusted instructions
  • Auto-run: CI/CD pipelines, fully automated workflows, stable instructions

Next

Agent handoff

Understand how captures are routed to your agent through MCP.

Configuration

All environment variables and setup options.