
Video, network traffic, console output and the symbolicated crash thread from the session that actually failed. Captured on device, waiting in the report. Nothing to reproduce.
Various data streams recorded in a rolling buffer on the device. When something fails, they arrive together.
Pixel-accurate replay of what the user actually saw, frame for frame.
Every request and response, inspectable down to headers, payload and timing.
System console plus your own log lines, in true chronological order.
Symbolicated stack traces, sitting next to the recording that produced them.
Taps, gestures and the active screen, overlaid on the video.
Mark views and fields as private and the pixels never leave the device.
A read of the evidence with a likely root cause, attached to the report.
Users shake to report, and the recording is already attached.
Scrub back through the seconds before failure. Video, console, requests and the crash thread all move on the same clock, so cause and effect line up instead of being guessed at.
Add the SDK, paste a token, ship. The same call on every platform — no build step, no proxy, no instrumenting your views.
The MCP server gives any MCP-aware agent scoped access to crashes, errors, builds and vulnerability scans — from inside the editor. Inference stays in your environment; no source code leaves your machine.
Read the MCP documentation →No agent to wire up: every captured signal feeds an analysis that names the likely cause and proposes the fix, inline in the issue.
Force-unwrap of data.orderId when /v1/checkout returned 500.
Drop in the SDK today and the next failure arrives with its own evidence.