Replaying recordings

What exists today

There is no standalone simdrive replay CLI. The engine’s console-script entry points are simdrive and simdrive-mcp (both start the MCP server) — see [project.scripts] in pyproject.toml. Replay runs through the MCP replay tool, from inside an agent session:

You (in Cursor / Claude Code):
  "Replay the ENG-1247-repro recording against iPhone 17 / iOS 26.3."

Claude:
  → replay({ name: "ENG-1247-repro", device: "iPhone 17" })

The replay itself doesn’t call a model — it re-observes the screen, resolves the recorded target, and checks the rendered frame against the recording via SSIM. What costs a model call is having an agent session available to invoke the replay tool in the first place.

Running it unattended (e.g. in CI)

Today that means running an MCP-capable agent (Claude Code or another MCP client) inside your CI job, wired to call replay for the recordings you want to check. There is no packaged GitHub Action, no headless replay runner, and no license required — SimDrive is free to install and use.

Coming soon

A dedicated CI story (headless replay invocation without a full agent session, matrix builds across simulator + real device, failure-artifact upload) is not built yet. This page will be updated when it ships.