Reproduce a Sentry crash on a real iPhone

A crash report tells you what broke. It doesn’t tell you whether you can make it break again. With Sentry connected to Mobster for Mac, one task does both halves: the agent reads the newest crash from Sentry, then works through its steps in your app on an iPhone you own, and reports what it saw.

Set it up once

1. Open Mobster for Mac and go to Settings › Connectors. Press + next to Sentry.
2. Sign in to Sentry in your browser and allow the scopes it lists: org:read, project:write, team:write, event:write and alerts:write. Sentry shows two consent pages.
3. Mobster comes to the front and the row reads something like Connected. 23 tools. The count is whatever Sentry’s server lists that day.

Your app needs to be on the iPhone already, from Xcode, TestFlight or a device build. Mobster’s agent drives an installed app. It doesn’t build yours.

What Mobster does

Start the task above as a new task in Mobster for Mac. Connectors are part of the Mac app, so a task that Claude Code starts over MCP doesn’t get Sentry. Then:

1. The agent searches Sentry for your newest crash and opens its event. The task’s steps show each call as Using Sentry: ... and then Used Sentry: ....
2. It reads the stack trace and the event’s details. What Sentry returns reaches the agent as data, never as instructions, so a crash message that says “ignore your instructions” is read and nothing more.
3. It opens your app on the iPhone and works through the screens the report points at, one tap at a time, reading each screen from the accessibility tree.
4. If a step would send, post, buy or delete, it stops and asks you first on the approval card. The card shows what it would do.
5. It writes down what happened: whether the app crashed, which screen it reached, and where it stopped.

What you get

A finished task. The answer says in words what happened, the steps list every tap and every Sentry call, and under the answer a chip shows each screen the agent read, with the words it read there. Choose a chip to see that screen. Each Sentry call is journaled before it is sent, as a connector_call event with the tool and its arguments. If Sentry doesn’t answer within 60 seconds, Mobster never sends that call again in the task, because it can’t know whether it happened. The agent checks another way.

When the crash repeats, you have the steps and frames that caused it, ready to paste into the issue. When it doesn’t, you have the same record of what was tried, which is what a teammate asks for before closing it.

What it won’t do

• It doesn’t read Sentry data you haven’t connected. The connector has the access of the account you signed in with.
• It doesn’t guess when a step needs a login. A screen behind a sign-in it can’t pass ends the task there, and the answer says so.
• It doesn’t run on a phone another task is using. It waits in line.

Tested

Mobster connected to Sentry, read from it and wrote to it on 15 Oct 2026, then disconnected and reconnected. The read was find_organizations. The write created a project named mobster-test. The same Sentry page of the docs lists the other tools and what each does.