Launch HN: Minicor (YC P26) – Windows desktop automations at scale

Faiz and Saheed didn't set out to build an RPA company. They were working on API integrations when a customer dangled a two-day deal if they could hook into a clinic's Windows-based medical records system, a piece of software with zero API access. That one favor turned into Minicor, a YC P26 startup betting that desktop automation is a much bigger, much harder problem than most people realize.

Anyone who's touched robotic process automation knows the pain. Scripting means learning quirky legacy UIs that shift without warning. Orchestration means babysitting virtual machines, queuing jobs, and figuring out how to run things in parallel without everything falling over. And debugging is close to impossible when there's no real observability into what actually happened during a run. Faiz and Saheed cite failure rates north of 30 percent as common in this world, with some companies generating thousands of support tickets a month just from broken automations quietly failing in production.

Minicor's fix is to let Claude Code or Codex drive an MCP server that controls a virtual machine and writes the automation as an actual Python script rather than a fragile sequence of recorded clicks. That matters because scripts are fast, cheap to run, and deterministic in a way that live agent-driven clicking never is. Every workflow can be triggered by API with a defined input and output schema, and every run gets logged with a video replay attached, so when something breaks, someone can actually see why instead of guessing.

The platform also handles the unglamorous stuff that tends to sink RPA projects at scale: cloning VMs so jobs can run in parallel, managing 2FA and OTP prompts that would otherwise require a human to babysit a login screen, and letting the underlying MCP agent go back in and patch broken code itself, with every change tracked in version control. Because the workflows are just code, the founders were able to bolt on Slack alerts, human-in-the-loop checkpoints, and even LLM-based sanity checks that look at a screenshot and judge whether the VM is in the expected state.

It's a narrow niche, automating software that refuses to modernize, but it's also one that isn't going away anytime soon. Healthcare records, legacy finance tools, government portals: plenty of critical systems are never getting a proper API, and someone has to keep the automations pointed at them from quietly rotting.