Troubleshooting

Captures come back blank or redacted

If your captures are solid black, empty, or have whole windows blanked out, Noru doesn't have the Screen Recording permission. macOS gates screen capture behind it, and without the grant it hands back an empty or redacted image, so a capture looks broken when really it's just blind.

1. Open System Settings, Privacy and Security, Screen Recording.
2. Turn on Noru Flow.
3. Quit and relaunch Noru so macOS picks up the change. This part matters, the permission doesn't take hold until you restart the app.

The first-run welcome card asks for this, but it's easy to skip or to lose when you move to a new Mac. Re-granting and relaunching fixes it every time.

The agent searches my repo for an image instead of pulling

You take a capture, ask your agent about it, and instead of reading your screen it goes rummaging through your project for an icon or an asset file. That's Cursor or VS Code Copilot without the rule. Claude Code reads the tool description and pulls on its own, but Cursor and Copilot don't, so left alone they assume "this image" lives in your files.

The fix is the one-time rule that tells them to call Noru first. Paste it into your global User Rules, not a single project, then reload the session. The exact text and where it goes are in the Cursor and VS Code rule. Once it's in, "fix this" pulls the capture every time.

My agent doesn't see Noru's tools at all

Agents read their MCP config once, at startup. If you connected Noru while a Claude Code or Cursor session was already open, that session never saw it. Restart the agent, or open a fresh session, and the Noru tools appear. New sessions always start with them.

If you're not sure the connection landed, re-run the installer. It's safe to repeat and only refreshes Noru's entry without touching your other servers:

noru-mcp install

If it reports that the claude command isn't on your PATH, it prints the exact line to run by hand. Everything else registers without you doing anything. The full walkthrough is in connect your agent.

The agent says the capture was "already delivered"

The live handoff, get_latest_capture, is consume-on-pull: it hands your most recent capture to whichever agent asks next, once. That's deliberate, it keeps a screenshot from an hour ago out of an unrelated prompt. So if you ask twice, or a second agent asks for the same one, the second call has nothing fresh to give.

This almost never needs a fix, it's working as intended. When you do want a capture again, it isn't gone, just no longer the "latest" handoff. Ask for it by history instead:

In practice you just say "the one before that" or "the screenshot of the other repo," and the agent reaches for the right tool. The tool reference spells out which is which.